Publikationen

Wissenschaftliche Publikationen des Instituts der letzten zwei Jahre

2023

Marked Neural Spatio-Temporal Point Process Involving a Dynamic Graph Neural Network. Moallemy-Oureh, Alice; Beddar-Wiesing, Silvia; Nather, Rüdiger; Thomas, Josephine. In Workshop on Temporal Graph Learning (TGL), NeurIPS, bll 1–7. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Temporal Point Processes (TPPs) have recently become increasingly interesting for learning dynamics in graph data. A reason for this is that learning on dynamic graph data is becoming more relevant, since data from many scientific fields, ranging from mathematics, biology, social sciences, and physics to computer science, is naturally related and inherently dynamic. In addition, TPPs provide a meaningful characterization of event streams and a prediction mechanism for future events. Therefore, (semi-)parameterized Neural TPPs have been introduced whose characterization can be (partially) learned and, thus, enable the representation of more complex phenomena. However, the research on modeling dynamic graphs with TPPs is relatively young, and only a few models for node attribute changes or evolving edges have been proposed yet. To allow for learning on fully dynamic graph streams, i.e., graphs that can change in their structure (addition/deletion of nodes/edge) and in their node/edge attributes, we propose a Marked Neural Spatio-Temporal Point Process (MNSTPP). It leverages a Dynamic Graph Neural Network to learn a Marked TPP that handles attributes and spatial data to model and predict any event in a graph stream.
@inproceedings{moallemyoureh2023marked, abstract = {Temporal Point Processes (TPPs) have recently become increasingly interesting for learning dynamics in graph data. A reason for this is that learning on dynamic graph data is becoming more relevant, since data from many scientific fields, ranging from mathematics, biology, social sciences, and physics to computer science, is naturally related and inherently dynamic. In addition, TPPs provide a meaningful characterization of event streams and a prediction mechanism for future events. Therefore, (semi-)parameterized Neural TPPs have been introduced whose characterization can be (partially) learned and, thus, enable the representation of more complex phenomena. However, the research on modeling dynamic graphs with TPPs is relatively young, and only a few models for node attribute changes or evolving edges have been proposed yet. To allow for learning on fully dynamic graph streams, i.e., graphs that can change in their structure (addition/deletion of nodes/edge) and in their node/edge attributes, we propose a Marked Neural Spatio-Temporal Point Process (MNSTPP). It leverages a Dynamic Graph Neural Network to learn a Marked TPP that handles attributes and spatial data to model and predict any event in a graph stream.}, author = {Moallemy-Oureh, Alice and Beddar-Wiesing, Silvia and Nather, Rüdiger and Thomas, Josephine}, booktitle = {Workshop on Temporal Graph Learning (TGL), NeurIPS}, keywords = {isac-www}, pages = {1–7}, title = {Marked Neural Spatio-Temporal Point Process Involving a Dynamic Graph Neural Network}, year = 2023 }
Power flow forecasts at transmission grid nodes using Graph Neural Networks. Beinert, Dominik; Holzhüter, Clara; Thomas, Josephine; Vogt, Stephan. In Energy and AI, 14(1), bl 100262. Elsevier, 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
The increasing share of renewable energy in the electricity grid and progressing changes in power consumption have led to fluctuating, and weather-dependent power flows. To ensure grid stability, grid operators rely on power forecasts which are crucial for grid calculations and planning. In this paper, a Multi-Task Learning approach is combined with a Graph Neural Network (GNN) to predict vertical power flows at transformers connecting high and extra-high voltage levels. The proposed method accounts for local differences in power flow characteristics by using an Embedding Multi-Task Learning approach. The use of a Bayesian embedding to capture the latent node characteristics allows to share the weights across all transformers in the subsequent node-invariant GNN while still allowing the individual behavioral patterns of the transformers to be distinguished. At the same time, dependencies between transformers are considered by the GNN architecture which can learn relationships between different transformers and thus take into account that power flows in an electricity network are not independent from each other. The effectiveness of the proposed method is demonstrated through evaluation on two real-world data sets provided by two of four German Transmission System Operators, comprising large portions of the operated German transmission grid. The results show that the proposed Multi-Task Graph Neural Network is a suitable representation learner for electricity networks with a clear advantage provided by the preceding embedding layer. It is able to capture interconnections between correlated transformers and indeed improves the performance in power flow prediction compared to standard Neural Networks. A sign test shows that the proposed model reduces the test RMSE on both data sets compared to the benchmark models significantly.
@article{beinert2023power, abstract = {The increasing share of renewable energy in the electricity grid and progressing changes in power consumption have led to fluctuating, and weather-dependent power flows. To ensure grid stability, grid operators rely on power forecasts which are crucial for grid calculations and planning. In this paper, a Multi-Task Learning approach is combined with a Graph Neural Network (GNN) to predict vertical power flows at transformers connecting high and extra-high voltage levels. The proposed method accounts for local differences in power flow characteristics by using an Embedding Multi-Task Learning approach. The use of a Bayesian embedding to capture the latent node characteristics allows to share the weights across all transformers in the subsequent node-invariant GNN while still allowing the individual behavioral patterns of the transformers to be distinguished. At the same time, dependencies between transformers are considered by the GNN architecture which can learn relationships between different transformers and thus take into account that power flows in an electricity network are not independent from each other. The effectiveness of the proposed method is demonstrated through evaluation on two real-world data sets provided by two of four German Transmission System Operators, comprising large portions of the operated German transmission grid. The results show that the proposed Multi-Task Graph Neural Network is a suitable representation learner for electricity networks with a clear advantage provided by the preceding embedding layer. It is able to capture interconnections between correlated transformers and indeed improves the performance in power flow prediction compared to standard Neural Networks. A sign test shows that the proposed model reduces the test RMSE on both data sets compared to the benchmark models significantly.}, author = {Beinert, Dominik and Holzhüter, Clara and Thomas, Josephine and Vogt, Stephan}, journal = {Energy and AI}, keywords = {isac-www}, number = 1, pages = 100262, publisher = {Elsevier}, title = {Power flow forecasts at transmission grid nodes using Graph Neural Networks}, volume = 14, year = 2023 }
Multi-annotator Deep Learning: A Probabilistic Framework for Classification. Herde, Marek; Huseljic, Denis; Sick, Bernhard. In Transactions on Machine Learning Research. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Solving complex classification tasks using deep neural networks typically requires large amounts of annotated data. However, corresponding class labels are noisy when provided by error-prone annotators, e.g., crowdworkers. Training standard deep neural networks leads to subpar performances in such multi-annotator supervised learning settings. We address this issue by presenting a probabilistic training framework named multi-annotator deep learning (MaDL). A downstream ground truth and an annotator performance model are jointly trained in an end-to-end learning approach. The ground truth model learns to predict instances' true class labels, while the annotator performance model infers probabilistic estimates of annotators' performances. A modular network architecture enables us to make varying assumptions regarding annotators' performances, e.g., an optional class or instance dependency. Further, we learn annotator embeddings to estimate annotators' densities within a latent space as proxies of their potentially correlated annotations. Together with a weighted loss function, we improve the learning from correlated annotation patterns. In a comprehensive evaluation, we examine three research questions about multi-annotator supervised learning. Our findings show MaDL's state-of-the-art performance and robustness against many correlated, spamming annotators.
@article{herde2023multi, abstract = {Solving complex classification tasks using deep neural networks typically requires large amounts of annotated data. However, corresponding class labels are noisy when provided by error-prone annotators, e.g., crowdworkers. Training standard deep neural networks leads to subpar performances in such multi-annotator supervised learning settings. We address this issue by presenting a probabilistic training framework named multi-annotator deep learning (MaDL). A downstream ground truth and an annotator performance model are jointly trained in an end-to-end learning approach. The ground truth model learns to predict instances' true class labels, while the annotator performance model infers probabilistic estimates of annotators' performances. A modular network architecture enables us to make varying assumptions regarding annotators' performances, e.g., an optional class or instance dependency. Further, we learn annotator embeddings to estimate annotators' densities within a latent space as proxies of their potentially correlated annotations. Together with a weighted loss function, we improve the learning from correlated annotation patterns. In a comprehensive evaluation, we examine three research questions about multi-annotator supervised learning. Our findings show MaDL's state-of-the-art performance and robustness against many correlated, spamming annotators.}, author = {Herde, Marek and Huseljic, Denis and Sick, Bernhard}, journal = {Transactions on Machine Learning Research}, keywords = {Crowdsourcing}, title = {Multi-annotator Deep Learning: A Probabilistic Framework for Classification}, year = 2023 }
Domain Imaging in Periodic Submicron Wide Nanostructures by Digital Drift Correction in Kerr Microscopy. Akhundzada, Sapida; Dingel, Kristina; Bischof, David; Janzen, Christian; Sick, Bernhard; Ehresmann, Arno. In Advanced Photonics Research, 4(10), bl 2300170. Wiley, 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Magneto-optical Kerr microscopy is a powerful method for imaging magnetic domains. Even though domain imaging below the diffractive resolution limit is possible, such investigations are getting increasingly complex with decreasing structure size due to the decreasing Kerr contrast. As magnetic domain images free of topographical artifacts are obtained by subtracting a reference image from the actual image, the corresponding challenges are additionally increased by unavoidable sample motion in the time interval between acquiring the two images. Software-based drift corrections typically rely on a unique structure in the image's region of interest (ROI), recognized automatically or selected manually by the user. By digital image shifting, the ROI positions in the actual and reference images are aligned, and the sample motion is compensated. For magnetic domain imaging in periodically arranged micro- or nano-objects, unique topographical features are not given, making the drift correction by ROIs difficult, often even impossible. Herein, a novel software-based approach is presented for drift corrections to image domains with features close/below the optical resolution limit and for investigating periodically arranged micro- or nano-objects without utilizing ROIs. High-contrast images are obtained, enabling the characterization of periodically arranged 1D, 2D, and 3D magnetic objects with lateral dimensions below 100 nm.
@article{akhundzada2023domain, abstract = {Magneto-optical Kerr microscopy is a powerful method for imaging magnetic domains. Even though domain imaging below the diffractive resolution limit is possible, such investigations are getting increasingly complex with decreasing structure size due to the decreasing Kerr contrast. As magnetic domain images free of topographical artifacts are obtained by subtracting a reference image from the actual image, the corresponding challenges are additionally increased by unavoidable sample motion in the time interval between acquiring the two images. Software-based drift corrections typically rely on a unique structure in the image's region of interest (ROI), recognized automatically or selected manually by the user. By digital image shifting, the ROI positions in the actual and reference images are aligned, and the sample motion is compensated. For magnetic domain imaging in periodically arranged micro- or nano-objects, unique topographical features are not given, making the drift correction by ROIs difficult, often even impossible. Herein, a novel software-based approach is presented for drift corrections to image domains with features close/below the optical resolution limit and for investigating periodically arranged micro- or nano-objects without utilizing ROIs. High-contrast images are obtained, enabling the characterization of periodically arranged 1D, 2D, and 3D magnetic objects with lateral dimensions below 100 nm.}, author = {Akhundzada, Sapida and Dingel, Kristina and Bischof, David and Janzen, Christian and Sick, Bernhard and Ehresmann, Arno}, journal = {Advanced Photonics Research}, keywords = {isac-www}, number = 10, pages = 2300170, publisher = {Wiley}, title = {Domain Imaging in Periodic Submicron Wide Nanostructures by Digital Drift Correction in Kerr Microscopy}, volume = 4, year = 2023 }
Unraveling the Complexity of Splitting Sequential Data: Tackling Challenges in Video and Time Series Analysis. Botache, Diego; Dingel, Kristina; Huhnstock, Rico; Ehresmann, Arno; Sick, Bernhard. In arXiv e-prints, bl arXiv:2307.14294. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Splitting of sequential data, such as videos and time series, is an essential step in various data analysis tasks, including object tracking and anomaly detection. However, splitting sequential data presents a variety of challenges that can impact the accuracy and reliability of subsequent analyses. This concept article examines the challenges associated with splitting sequential data, including data acquisition, data representation, split ratio selection, setting up quality criteria, and choosing suitable selection strategies. We explore these challenges through two real-world examples: motor test benches and particle tracking in liquids.
@article{botache2023unraveling, abstract = {Splitting of sequential data, such as videos and time series, is an essential step in various data analysis tasks, including object tracking and anomaly detection. However, splitting sequential data presents a variety of challenges that can impact the accuracy and reliability of subsequent analyses. This concept article examines the challenges associated with splitting sequential data, including data acquisition, data representation, split ratio selection, setting up quality criteria, and choosing suitable selection strategies. We explore these challenges through two real-world examples: motor test benches and particle tracking in liquids.}, author = {Botache, Diego and Dingel, Kristina and Huhnstock, Rico and Ehresmann, Arno and Sick, Bernhard}, journal = {arXiv e-prints}, keywords = {Sensor-Data}, pages = {arXiv:2307.14294}, title = {Unraveling the Complexity of Splitting Sequential Data: Tackling Challenges in Video and Time Series Analysis}, year = 2023 }
The IMPTC Dataset: An Infrastructural Multi-Person Trajectory and Context Dataset. Hetzel, Manuel; Reichert, Hannes; Reitberger, Günther; Doll, Konrad; Sick, Bernhard; Fuchs, Erich. In IEEE Intelligent Vehicles Symposium (IV), bll 1–7. IEEE, 2023.
- [ Abstract ]
- [ BibTeX ]
Inner-city intersections are among the most critical traffic areas for injury and fatal accidents. Automated vehicles struggle with the complex and hectic everyday life within those areas. Sensor-equipped smart infrastructures, which can cooperate with vehicles, can benefit automated traffic by extending the perception capabilities of drivers and vehicle perception systems. Additionally, they offer the opportunity to gather reproducible and precise data of a holistic scene understanding, including context information as a basis for training algorithms for various applications in automated traffic. Therefore, we introduce the Infrastructural Multi-Person Trajectory and Context Dataset (IMPTC). We use an intelligent public inner-city intersection in Germany with visual sensor technology. A multi-view camera and LiDAR system perceives traffic situations and road users’ behavior. Additional sensors monitor contextual information like weather, lighting, and traffic light signal status. The data acquisition system focuses on Vulnerable Road Users (VRUs) and multiagent interaction. The resulting dataset consists of eight hours of measurement data. It contains over 2,500 VRU trajectories, including pedestrians, cyclists, e-scooter riders, strollers, and wheelchair users, and over 20,000 vehicle trajectories at different day times, weather conditions, and seasons. In addition, to enable the entire stack of research capabilities, the dataset includes all data, starting from the sensor-, calibration- and detection data until trajectory and context data. The dataset is continuously expanded and is available online for non-commercial research at https://github.com/kav-institute/imptc-dataset
@inproceedings{hetzel2023imptc, abstract = {Inner-city intersections are among the most critical traffic areas for injury and fatal accidents. Automated vehicles struggle with the complex and hectic everyday life within those areas. Sensor-equipped smart infrastructures, which can cooperate with vehicles, can benefit automated traffic by extending the perception capabilities of drivers and vehicle perception systems. Additionally, they offer the opportunity to gather reproducible and precise data of a holistic scene understanding, including context information as a basis for training algorithms for various applications in automated traffic. Therefore, we introduce the Infrastructural Multi-Person Trajectory and Context Dataset (IMPTC). We use an intelligent public inner-city intersection in Germany with visual sensor technology. A multi-view camera and LiDAR system perceives traffic situations and road users’ behavior. Additional sensors monitor contextual information like weather, lighting, and traffic light signal status. The data acquisition system focuses on Vulnerable Road Users (VRUs) and multiagent interaction. The resulting dataset consists of eight hours of measurement data. It contains over 2,500 VRU trajectories, including pedestrians, cyclists, e-scooter riders, strollers, and wheelchair users, and over 20,000 vehicle trajectories at different day times, weather conditions, and seasons. In addition, to enable the entire stack of research capabilities, the dataset includes all data, starting from the sensor-, calibration- and detection data until trajectory and context data. The dataset is continuously expanded and is available online for non-commercial research at https://github.com/kav-institute/imptc-dataset}, author = {Hetzel, Manuel and Reichert, Hannes and Reitberger, Günther and Doll, Konrad and Sick, Bernhard and Fuchs, Erich}, booktitle = {IEEE Intelligent Vehicles Symposium (IV)}, keywords = {isac-www}, pages = {1–7}, publisher = {IEEE}, title = {The {IMPTC} Dataset: An Infrastructural Multi-Person Trajectory and Context Dataset}, year = 2023 }
Sensor Equivariance by LiDAR Projection Images. Reichert, Hannes; Hetzel, Manuel; Schreck, Steven; Doll, Konrad; Sick, Bernhard. In IEEE Intelligent Vehicles Symposium (IV), bll 1–6. IEEE, 2023.
- [ Abstract ]
- [ BibTeX ]
In this work, we propose an extension of conventional image data by an additional channel in which the associated projection properties are encoded. This addresses the issue of sensor-dependent object representation in projection-based sensors, such as LiDAR, which can lead to distorted physical and geometric properties due to variations in sensor resolution and field of view. To that end, we propose an architecture for processing this data in an instance segmentation framework. We focus specifically on LiDAR as a key sensor modality for machine vision tasks and highly automated driving (HAD). Through an experimental setup in a controlled synthetic environment, we identify a bias on sensor resolution and field of view and demonstrate that our proposed method can reduce said bias for the task of LiDAR instance segmentation. Furthermore, we define our method such that it can be applied to other projection-based sensors, such as cameras. To promote transparency, we make our code and dataset publicly available. This method shows the potential to improve performance and robustness in various machine vision tasks that utilize projection-based sensors.
@inproceedings{reichert2023sensor, abstract = {In this work, we propose an extension of conventional image data by an additional channel in which the associated projection properties are encoded. This addresses the issue of sensor-dependent object representation in projection-based sensors, such as LiDAR, which can lead to distorted physical and geometric properties due to variations in sensor resolution and field of view. To that end, we propose an architecture for processing this data in an instance segmentation framework. We focus specifically on LiDAR as a key sensor modality for machine vision tasks and highly automated driving (HAD). Through an experimental setup in a controlled synthetic environment, we identify a bias on sensor resolution and field of view and demonstrate that our proposed method can reduce said bias for the task of LiDAR instance segmentation. Furthermore, we define our method such that it can be applied to other projection-based sensors, such as cameras. To promote transparency, we make our code and dataset publicly available. This method shows the potential to improve performance and robustness in various machine vision tasks that utilize projection-based sensors.}, author = {Reichert, Hannes and Hetzel, Manuel and Schreck, Steven and Doll, Konrad and Sick, Bernhard}, booktitle = {IEEE Intelligent Vehicles Symposium (IV)}, keywords = {isac-www}, pages = {1–6}, publisher = {IEEE}, title = {Sensor Equivariance by {LiDAR} Projection Images}, year = 2023 }
Managing power grids through topology actions: A comparative study between advanced rule-based and reinforcement learning agents. Lehna, Malte; Viebahn, Jan; Marot, Antoine; Tomforde, Sven; Scholz, Christoph. In Energy and AI, 14, bl 100276. Energy and AI, 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
The operation of electricity grids has become increasingly complex due to the current upheaval and the increase in renewable energy production. As a consequence, active grid management is reaching its limits with conventional approaches. In the context of the Learning to Run a Power Network (L2RPN) challenge, it has been shown that Reinforcement Learning (RL) is an efficient and reliable approach with considerable potential for automatic grid operation. In this article, we analyse the submitted agent from Binbinchen and provide novel strategies to improve the agent, both for the RL and the rule-based approach. The main improvement is a N-1 strategy, where we consider topology actions that keep the grid stable, even if one line is disconnected. More, we also propose a topology reversion to the original grid, which proved to be beneficial. The improvements are tested against reference approaches on the challenge test sets and are able to increase the performance of the rule-based agent by 27%. In direct comparison between rule-based and RL agent we find similar performance. However, the RL agent has a clear computational advantage. We also analyse the behaviour in an exemplary case in more detail to provide additional insights. Here, we observe that through the N-1 strategy, the actions of both the rule-based and the RL agent become more diversified.
@article{lehna2023managing, abstract = {The operation of electricity grids has become increasingly complex due to the current upheaval and the increase in renewable energy production. As a consequence, active grid management is reaching its limits with conventional approaches. In the context of the Learning to Run a Power Network (L2RPN) challenge, it has been shown that Reinforcement Learning (RL) is an efficient and reliable approach with considerable potential for automatic grid operation. In this article, we analyse the submitted agent from Binbinchen and provide novel strategies to improve the agent, both for the RL and the rule-based approach. The main improvement is a N-1 strategy, where we consider topology actions that keep the grid stable, even if one line is disconnected. More, we also propose a topology reversion to the original grid, which proved to be beneficial. The improvements are tested against reference approaches on the challenge test sets and are able to increase the performance of the rule-based agent by 27%. In direct comparison between rule-based and RL agent we find similar performance. However, the RL agent has a clear computational advantage. We also analyse the behaviour in an exemplary case in more detail to provide additional insights. Here, we observe that through the N-1 strategy, the actions of both the rule-based and the RL agent become more diversified.}, author = {Lehna, Malte and Viebahn, Jan and Marot, Antoine and Tomforde, Sven and Scholz, Christoph}, journal = {Energy and AI}, keywords = {ProximalPolicyOptimisation}, pages = 100276, publisher = {Energy and AI}, title = {Managing power grids through topology actions: A comparative study between advanced rule-based and reinforcement learning agents}, volume = 14, year = 2023 }
A Unified Approach to Communication Delay and Communication Frequency in Distributed State Estimation of Linear Systems. Mast, J.; Liu, Z.; Wang, Z.; Stursberg, O. In IEEE Control Systems Letters, Volume 7, bll 2755–2760. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This letter introduces a novel method to address communication delay and communication frequency in distributed state estimation using a unified structure. Starting from a general type of delay, which can be time-varying, unknown, and different for each channel, two estimation strategies are proposed which use either a unified upper-bound of the delay in all channels, or a sender specific upper-bound in order to improve the performance. The strategies guarantee convergence of the estimation error, and they can be applied to study how communication frequency (or asynchronous communication) affects the convergence rate of the estimation. Thus, the task of determining a necessary communication frequency for obtaining specified performance of distributed estimation is considered, and this task is approached as a special case of the one tailored to delay. Effectiveness of the techniques is confirmed by numerical examples for both types of tasks.
@article{mast2023unified, abstract = {This letter introduces a novel method to address communication delay and communication frequency in distributed state estimation using a unified structure. Starting from a general type of delay, which can be time-varying, unknown, and different for each channel, two estimation strategies are proposed which use either a unified upper-bound of the delay in all channels, or a sender specific upper-bound in order to improve the performance. The strategies guarantee convergence of the estimation error, and they can be applied to study how communication frequency (or asynchronous communication) affects the convergence rate of the estimation. Thus, the task of determining a necessary communication frequency for obtaining specified performance of distributed estimation is considered, and this task is approached as a special case of the one tailored to delay. Effectiveness of the techniques is confirmed by numerical examples for both types of tasks.}, author = {Mast, J. and Liu, Z. and Wang, Z. and Stursberg, O.}, journal = {IEEE Control Systems Letters}, keywords = {isac-www}, pages = {2755-2760}, title = {A Unified Approach to Communication Delay and Communication Frequency in Distributed State Estimation of Linear Systems}, volume = {Volume 7}, year = 2023 }
Model Predictive Control of PDEs for Temperature Control in 3D-Printing Processes. Schmidtke, V.; Rüger, M.; Stursberg, O. In Proc. of the 2023 European Control Conference, bll 496–501. 2023.
- [ Abstract ]
- [ BibTeX ]
Processes of additive manufacturing by 3D-printing are typically operated by only using constant references for feed motion and laser power. For adapting properties of workpieces to desired geometries, the control towards changing reference signals is an enabling factor, though. With a focus on temperature control, this paper proposes a scheme of model predictive control for 3D-printing with consideration of constraints and arbitrary references. Starting from partial differential equations to model heat propagation over the geometry, prediction models and optimization problems are formulated as a local part of a more detailed simulation model of the complete workpiece. For laser velocity and power as manipulated variables, optimization problems involving integer variables are solved. Simulation results show how spatial temperature profiles can be adjusted by the proposed approach.
@article{schmidtke2023model, abstract = {Processes of additive manufacturing by 3D-printing are typically operated by only using constant references for feed motion and laser power. For adapting properties of workpieces to desired geometries, the control towards changing reference signals is an enabling factor, though. With a focus on temperature control, this paper proposes a scheme of model predictive control for 3D-printing with consideration of constraints and arbitrary references. Starting from partial differential equations to model heat propagation over the geometry, prediction models and optimization problems are formulated as a local part of a more detailed simulation model of the complete workpiece. For laser velocity and power as manipulated variables, optimization problems involving integer variables are solved. Simulation results show how spatial temperature profiles can be adjusted by the proposed approach.}, author = {Schmidtke, V. and Rüger, M. and Stursberg, O.}, journal = {Proc. of the 2023 European Control Conference}, keywords = {isac-www}, pages = {496-501}, title = {Model Predictive Control of PDEs for Temperature Control in 3D-Printing Processes.}, year = 2023 }
On the Design of Limit Cycles of Planar Switching Affine Systems. Hanke, N.; Stursberg, O. In Proc. of the 2023 European Control Conference, bll 2251–2256. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
In the context of studying periodic processes, this paper investigates first under which conditions switching affine systems in the plane generate stable limit cycles. Based on these conditions, a design methodology is proposed by which the phase portraits of the switching systems are determined to obtain globally stable limit cycles from simple specifications, such as given amplitudes and frequencies of desired oscillations. As an application, the paper finally shows that an oscillator model can be derived with a small effort from data measured for an unknown oscillating system.
@article{nils2023design, abstract = {In the context of studying periodic processes, this paper investigates first under which conditions switching affine systems in the plane generate stable limit cycles. Based on these conditions, a design methodology is proposed by which the phase portraits of the switching systems are determined to obtain globally stable limit cycles from simple specifications, such as given amplitudes and frequencies of desired oscillations. As an application, the paper finally shows that an oscillator model can be derived with a small effort from data measured for an unknown oscillating system.}, author = {Hanke, N. and Stursberg, O.}, journal = {Proc. of the 2023 European Control Conference}, keywords = {isac-www}, pages = {2251-2256}, title = {On the Design of Limit Cycles of Planar Switching Affine Systems.}, year = 2023 }
Sampling-based Uncertainty Estimation for an Instance Segmentation Network. Heidecker, Florian; El-Khateeb, Ahmad; Sick, Bernhard. In arXiv e-prints, bl arXiv:2305.14977. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
The examination of uncertainty in the predictions of machine learning (ML) models is receiving increasing attention. One uncertainty modeling technique used for this purpose is Monte-Carlo (MC)-Dropout, where repeated predictions are generated for a single input. Therefore, clustering is required to describe the resulting uncertainty, but only through efficient clustering is it possible to describe the uncertainty from the model attached to each object. This article uses Bayesian Gaussian Mixture (BGM) to solve this problem. In addition, we investigate different values for the dropout rate and other techniques, such as focal loss and calibration, which we integrate into the Mask-RCNN model to obtain the most accurate uncertainty approximation of each instance and showcase it graphically.
@article{heidecker2023sampling, abstract = {The examination of uncertainty in the predictions of machine learning (ML) models is receiving increasing attention. One uncertainty modeling technique used for this purpose is Monte-Carlo (MC)-Dropout, where repeated predictions are generated for a single input. Therefore, clustering is required to describe the resulting uncertainty, but only through efficient clustering is it possible to describe the uncertainty from the model attached to each object. This article uses Bayesian Gaussian Mixture (BGM) to solve this problem. In addition, we investigate different values for the dropout rate and other techniques, such as focal loss and calibration, which we integrate into the Mask-RCNN model to obtain the most accurate uncertainty approximation of each instance and showcase it graphically.}, author = {Heidecker, Florian and El-Khateeb, Ahmad and Sick, Bernhard}, journal = {arXiv e-prints}, keywords = {isac-www}, pages = {arXiv:2305.14977}, title = {Sampling-based Uncertainty Estimation for an Instance Segmentation Network}, year = 2023 }
Graph Neural Networks Designed for Different Graph Types: A Survey. Thomas, Josephine; Moallemy-Oureh, Alice; Beddar-Wiesing, Silvia; Holzhüter, Clara. In Transactions on Machine Learning Research. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Graphs are ubiquitous in nature and can therefore serve as models for many practical but also theoretical problems. For this purpose, they can be defined as many different types which suitably reflect the individual contexts of the represented problem. To address cutting-edge problems based on graph data, the research field of Graph Neural Networks (GNNs) has emerged. Despite the field’s youth and the speed at which new models are developed, many recent surveys have been published to keep track of them. Nevertheless, it has not yet been gathered which GNN can process what kind of graph types. In this survey, we give a detailed overview of already existing GNNs and, unlike previous surveys, categorize them according to their ability to handle different graph types and properties. We consider GNNs operating on static and dynamic graphs of different structural constitutions, with or without node or edge attributes. Moreover, we distinguish between GNN models for discrete-time or continuous-time dynamic graphs and group the models according to their architecture. We find that there are still graph types that are not or only rarely covered by existing GNN models. We point out where models are missing and give potential reasons for their absence.
@article{thomas2023graph, abstract = {Graphs are ubiquitous in nature and can therefore serve as models for many practical but also theoretical problems. For this purpose, they can be defined as many different types which suitably reflect the individual contexts of the represented problem. To address cutting-edge problems based on graph data, the research field of Graph Neural Networks (GNNs) has emerged. Despite the field’s youth and the speed at which new models are developed, many recent surveys have been published to keep track of them. Nevertheless, it has not yet been gathered which GNN can process what kind of graph types. In this survey, we give a detailed overview of already existing GNNs and, unlike previous surveys, categorize them according to their ability to handle different graph types and properties. We consider GNNs operating on static and dynamic graphs of different structural constitutions, with or without node or edge attributes. Moreover, we distinguish between GNN models for discrete-time or continuous-time dynamic graphs and group the models according to their architecture. We find that there are still graph types that are not or only rarely covered by existing GNN models. We point out where models are missing and give potential reasons for their absence.}, author = {Thomas, Josephine and Moallemy-Oureh, Alice and Beddar-Wiesing, Silvia and Holzhüter, Clara}, journal = {Transactions on Machine Learning Research}, keywords = {graph_neural_network}, title = {Graph Neural Networks Designed for Different Graph Types: A Survey}, year = 2023 }
DNN-Approximations of Control Laws for Nonlinear Systems with Polytopic Constraints. Markolf, L.; Stursberg, O. In Proc. of the 22nd IFAC World Congress, bll 1516–1521. 2023.
- [ Abstract ]
- [ BibTeX ]
This work focuses on deep artificial feed-forward neural networks as parametric function approximators in optimal control of discrete-time nonlinear but affine-in-control systems subject to polytopic constraints on the continuous states and control inputs. The neural networks are either considered as approximators of optimal control laws or as approximators of the corresponding cost functions. In both cases, an approach is developed for determining the approximated optimal control inputs without violating the constraints. A simple approximate policy iteration algorithm exploiting both approaches is finally presented and illustrated for a numerical example.
@article{markolf2023dnnapproximations, abstract = {This work focuses on deep artificial feed-forward neural networks as parametric function approximators in optimal control of discrete-time nonlinear but affine-in-control systems subject to polytopic constraints on the continuous states and control inputs. The neural networks are either considered as approximators of optimal control laws or as approximators of the corresponding cost functions. In both cases, an approach is developed for determining the approximated optimal control inputs without violating the constraints. A simple approximate policy iteration algorithm exploiting both approaches is finally presented and illustrated for a numerical example.}, author = {Markolf, L. and Stursberg, O.}, journal = {Proc. of the 22nd IFAC World Congress}, keywords = {isac-www}, pages = {1516-1521}, title = {DNN-Approximations of Control Laws for Nonlinear Systems with Polytopic Constraints.}, year = 2023 }
Exploring the Potential of Optimal Active Learning via a Non-myopic Oracle Policy. Sandrock, Christoph; Herde, Marek; Kottke, Daniel; Sick, Bernhard. In Discovery Science (DS), bll 265–276. Springer, 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Active learning aims to reduce the amount of labeled data while maximizing machine learning models’ performances. Currently, there is sparse research on the potential of an optimal active learning strategy. Therefore, we propose a non-myopic oracle policy that accesses the true labels of the data pool to approximate an optimal active learning strategy. We evaluate how the hyperparameters of this oracle policy influence its performance and empirically demonstrate that it is an upper baseline for common active learning strategies while being faster than a state-of-the-art oracle policy. For the sake of reproducibility, all the code related to our research is publicly available on our GitHub repository at https://github.com/ies-research/non-myopic-oracle-policy.
@inproceedings{sandrock2023exploring, abstract = {Active learning aims to reduce the amount of labeled data while maximizing machine learning models’ performances. Currently, there is sparse research on the potential of an optimal active learning strategy. Therefore, we propose a non-myopic oracle policy that accesses the true labels of the data pool to approximate an optimal active learning strategy. We evaluate how the hyperparameters of this oracle policy influence its performance and empirically demonstrate that it is an upper baseline for common active learning strategies while being faster than a state-of-the-art oracle policy. For the sake of reproducibility, all the code related to our research is publicly available on our GitHub repository at https://github.com/ies-research/non-myopic-oracle-policy.}, author = {Sandrock, Christoph and Herde, Marek and Kottke, Daniel and Sick, Bernhard}, booktitle = {Discovery Science (DS)}, keywords = {Classification}, pages = {265–276}, publisher = {Springer}, title = {Exploring the Potential of Optimal Active Learning via a Non-myopic Oracle Policy}, year = 2023 }
Tailored Output Layers of Neural Networks for Satisfaction of State Constraints in Nonlinear Control Systems. Markolf, L.; Stursberg, O. In Proceedings of the 2023 Annual American Control Conference (ACC), bll 1881–1888. 2023.
- [ Abstract ]
- [ BibTeX ]
This work considers the synthesis of state feedback controllers established as deep artificial feed-forward neural networks for the control of discrete-time nonlinear but input-affine systems. The idea is to design output layers of particular structure to guarantee the satisfaction of state constraints in form of control-invariant ellipsoids. Since an analytical expression can be derived for the resulting neural network controller, the latter can be stored and evaluated efficiently. Moreover, the proposed output layer guarantees the satisfaction of the considered state constraints for each specification of the parameter vector. Numerical examples are provided for illustration and evaluation of the approach, in which the approximation of a nonlinear model predictive control law is considered as application.
@article{markolf2023tailored, abstract = {This work considers the synthesis of state feedback controllers established as deep artificial feed-forward neural networks for the control of discrete-time nonlinear but input-affine systems. The idea is to design output layers of particular structure to guarantee the satisfaction of state constraints in form of control-invariant ellipsoids. Since an analytical expression can be derived for the resulting neural network controller, the latter can be stored and evaluated efficiently. Moreover, the proposed output layer guarantees the satisfaction of the considered state constraints for each specification of the parameter vector. Numerical examples are provided for illustration and evaluation of the approach, in which the approximation of a nonlinear model predictive control law is considered as application.}, author = {Markolf, L. and Stursberg, O.}, journal = {Proceedings of the 2023 Annual American Control Conference (ACC)}, keywords = {isac-www}, pages = {1881-1888}, title = {Tailored Output Layers of Neural Networks for Satisfaction of State Constraints in Nonlinear Control Systems}, year = 2023 }
Distributed Interval Estimation for Continuous-Time Linear Systems: A Two-Step Method. Wang, Z.; Zhang, J.; Liu, Z.; Han, W.; Shen, M. In Proc. of the 22nd IFAC World Congress, Volume 56(Issue 2), bll 8500–8505. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This paper proposes a two-step distributed interval estimation method for continuous-time linear time-invariant (LTI) systems subject to unknown but bounded disturbance and measurement noise. First, a network of local observers is designed to obtain point-wise estimation. Second, interval estimation is obtained based on the point-wise estimation and peak-to-peak analysis. The conditions of the observer design and the peak-to-peak analysis are formulated as linear matrix inequalities, which can be solved efficiently. Compared with the distributed interval observers designed using the monotone system theory, the proposed method has less restrictive design conditions and hence broader application scopes. Moreover, the parameter matrices in the proposed method can be optimized to obtain tight interval estimation. A numerical simulation is given to illustrate the performance of the proposed method.
@article{wang2023distributed, abstract = {This paper proposes a two-step distributed interval estimation method for continuous-time linear time-invariant (LTI) systems subject to unknown but bounded disturbance and measurement noise. First, a network of local observers is designed to obtain point-wise estimation. Second, interval estimation is obtained based on the point-wise estimation and peak-to-peak analysis. The conditions of the observer design and the peak-to-peak analysis are formulated as linear matrix inequalities, which can be solved efficiently. Compared with the distributed interval observers designed using the monotone system theory, the proposed method has less restrictive design conditions and hence broader application scopes. Moreover, the parameter matrices in the proposed method can be optimized to obtain tight interval estimation. A numerical simulation is given to illustrate the performance of the proposed method.}, author = {Wang, Z. and Zhang, J. and Liu, Z. and Han, W. and Shen, M.}, journal = {Proc. of the 22nd IFAC World Congress}, keywords = {isac-www}, number = {Issue 2}, pages = {8500-8505}, title = {Distributed Interval Estimation for Continuous-Time Linear Systems: A Two-Step Method.}, volume = {Volume 56}, year = 2023 }
On Optimal Synchronization of Diffusively Coupled Heterogeneous Van Der Pol Oscillators. Trummel, T.; Liu, Z.; Stursberg, O. In IFAC-PapersOnLine, Volume 56(Issue 2), bll 9475–9480. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This paper proposes a novel method to achieve and preserve synchronization for a set of connected heterogeneous Van der Pol oscillators. Unlike the state-of-the-art synchronization methods, in which a large coupling gain is applied to couple any pair of connected oscillators, the proposed method first casts the whole synchronization process into two phases. The first one considers the period from the beginning to the first instant of synchronization, while the second phase covers the following time in which synchronization must be preserved. It is shown that a large coupling gain is adopted for the first phase, while the averaged coupling gain to preserve the synchronization in the second phase can be reduced significantly by using an offline optimized coupling law. Efficiency and performance of this method are confirmed by a set of numerical tests with different graphs and system dynamics.
@article{trummel2023optimal, abstract = {This paper proposes a novel method to achieve and preserve synchronization for a set of connected heterogeneous Van der Pol oscillators. Unlike the state-of-the-art synchronization methods, in which a large coupling gain is applied to couple any pair of connected oscillators, the proposed method first casts the whole synchronization process into two phases. The first one considers the period from the beginning to the first instant of synchronization, while the second phase covers the following time in which synchronization must be preserved. It is shown that a large coupling gain is adopted for the first phase, while the averaged coupling gain to preserve the synchronization in the second phase can be reduced significantly by using an offline optimized coupling law. Efficiency and performance of this method are confirmed by a set of numerical tests with different graphs and system dynamics.}, author = {Trummel, T. and Liu, Z. and Stursberg, O.}, journal = {IFAC-PapersOnLine}, keywords = {isac-www}, number = {Issue 2}, pages = {9475-9480}, title = {On Optimal Synchronization of Diffusively Coupled Heterogeneous Van Der Pol Oscillators}, volume = {Volume 56}, year = 2023 }
Distributed MPC of Uncertain Multi-Agent Systems Considering Formations and Obstacles. Flüs, P.; Stursberg, O. In Proc. of the 22nd IFAC World Congress, Volume 56(Issue 2), bll 10155–10161. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This paper proposes a method to control a class of multi-agent systems with uncertainties modeled as stochastic processes with arbitrary probability distributions. The considered control problem is to lead a formation of agents through a space which is partially obstructed by obstacles. The proposed solution is to use a hierarchically structured approach of distributed stochastic model predictive control (DSMPC). The approach combines elements of formation reference structures, leader-follower concepts, and successive convexification (SC) for collision avoidance. To consider the stochastic uncertainties, over-approximated probabilistic reachable sets (PRS) are computed based on Chebyshev's inequality. The nominal (expected) agent behavior is optimized within the DSMPC such that (probabilistic) constraints are satisfied in a distributed way. For the overall approach, closed-loop stability of the distributed control concept is investigated and an illustrating example is provided.
@article{flus2023distributed, abstract = {This paper proposes a method to control a class of multi-agent systems with uncertainties modeled as stochastic processes with arbitrary probability distributions. The considered control problem is to lead a formation of agents through a space which is partially obstructed by obstacles. The proposed solution is to use a hierarchically structured approach of distributed stochastic model predictive control (DSMPC). The approach combines elements of formation reference structures, leader-follower concepts, and successive convexification (SC) for collision avoidance. To consider the stochastic uncertainties, over-approximated probabilistic reachable sets (PRS) are computed based on Chebyshev's inequality. The nominal (expected) agent behavior is optimized within the DSMPC such that (probabilistic) constraints are satisfied in a distributed way. For the overall approach, closed-loop stability of the distributed control concept is investigated and an illustrating example is provided.}, author = {Flüs, P. and Stursberg, O.}, journal = {Proc. of the 22nd IFAC World Congress}, keywords = {isac-www}, number = {Issue 2}, pages = {10155-10161}, title = {Distributed MPC of Uncertain Multi-Agent Systems Considering Formations and Obstacles}, volume = {Volume 56}, year = 2023 }
Reference Tracking for Constrained Uncertain Linear Systems by Stochastic MPC. Hahn, J.; Stursberg, O. In Proc. of the 22nd IFAC World Congress, 56(2), bll 10421–10427. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This paper addresses model predictive control of a class of linear systems subject to additive stochastic disturbances and constraints. The underlying stochastic optimal control problem combines inverse cumulative distribution functions with ellipsoid-in-polyhedron formulations to reduce the conservatism induced by constraint satisfaction. By use of terminal constraints and time-varying weights within the cost functional, the presented control scheme satisfies criteria for mean-square stability and can be adapted to reference-tracking problems for arbitrary reference signals.
@article{hahn2023reference, abstract = {This paper addresses model predictive control of a class of linear systems subject to additive stochastic disturbances and constraints. The underlying stochastic optimal control problem combines inverse cumulative distribution functions with ellipsoid-in-polyhedron formulations to reduce the conservatism induced by constraint satisfaction. By use of terminal constraints and time-varying weights within the cost functional, the presented control scheme satisfies criteria for mean-square stability and can be adapted to reference-tracking problems for arbitrary reference signals.}, author = {Hahn, J. and Stursberg, O.}, journal = {Proc. of the 22nd IFAC World Congress}, keywords = {isac-www}, number = 2, pages = {10421-10427}, title = {Reference Tracking for Constrained Uncertain Linear Systems by Stochastic MPC}, volume = 56, year = 2023 }
Reconstruction of incomplete X-ray diffraction pole figures of oligocrystalline materials using deep learning. Meier, David; Ragunathan, Rishan; Degener, Sebastian; Liehr, Alexander; Vollmer, Malte; Niendorf, Thomas; Sick, Bernhard. In Scientific Reports, 13(1), bl 5410. Springer Nature, 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
X‑ray diffraction crystallography allows non‑destructive examination of crystal structures. Furthermore, it has low requirements regarding surface preparation, especially compared to electron backscatter diffraction. However, up to now, X‑ray diffraction has been highly time‑ consuming in standard laboratory conditions since intensities on multiple lattice planes have to be recorded by rotating and tilting. Furthermore, examining oligocrystalline materials is challenging due to the limited number of diffraction spots. Moreover, commonly used evaluation methods for crystallographic orientation analysis need multiple lattice planes for a reliable pole figure reconstruction. In this article, we propose a deep‑learning‑based method for oligocrystalline specimens, i.e., specimens with up to three grains of arbitrary crystal orientations. Our approach allows faster experimentation due to accurate reconstructions of pole figure regions, which we did not probe experimentally. In contrast to other methods, the pole figure is reconstructed based on only a single incomplete pole figure. To speed up the development of our proposed method and for usage in other machine learning algorithms, we introduce a GPU‑based simulation for data generation. Furthermore, we present a pole widths standardization technique using a custom deep learning architecture that makes algorithms more robust against influences from the experiment setup and material.
@article{meier2023reconstruction, abstract = {X‑ray diffraction crystallography allows non‑destructive examination of crystal structures. Furthermore, it has low requirements regarding surface preparation, especially compared to electron backscatter diffraction. However, up to now, X‑ray diffraction has been highly time‑ consuming in standard laboratory conditions since intensities on multiple lattice planes have to be recorded by rotating and tilting. Furthermore, examining oligocrystalline materials is challenging due to the limited number of diffraction spots. Moreover, commonly used evaluation methods for crystallographic orientation analysis need multiple lattice planes for a reliable pole figure reconstruction. In this article, we propose a deep‑learning‑based method for oligocrystalline specimens, i.e., specimens with up to three grains of arbitrary crystal orientations. Our approach allows faster experimentation due to accurate reconstructions of pole figure regions, which we did not probe experimentally. In contrast to other methods, the pole figure is reconstructed based on only a single incomplete pole figure. To speed up the development of our proposed method and for usage in other machine learning algorithms, we introduce a GPU‑based simulation for data generation. Furthermore, we present a pole widths standardization technique using a custom deep learning architecture that makes algorithms more robust against influences from the experiment setup and material.}, author = {Meier, David and Ragunathan, Rishan and Degener, Sebastian and Liehr, Alexander and Vollmer, Malte and Niendorf, Thomas and Sick, Bernhard}, journal = {Scientific Reports}, keywords = {isac-www}, number = 1, pages = 5410, publisher = {Springer Nature}, title = {Reconstruction of incomplete {X}-ray diffraction pole figures of oligocrystalline materials using deep learning}, volume = 13, year = 2023 }
Utilizing Continuous Kernels for Processing Irregularly and Inconsistently Sampled Data With Position-Dependent Features. Magnussen, Birk Martin; Stern, Claudius; Sick, Bernhard. In International Conference on Autonomic and Autonomous Systems (ICAS), bll 49–53. ThinkMind, 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Continuous Kernels have been a recent development in convolutional neural networks. Such kernels are used to process data sampled at different resolutions as well as irregularly and inconsistently sampled data. Convolutional neural networks have the property of translational invariance (e.g., features are detected regardless of their position in the measurement domain), which is unsuitable for certain types of data, where the position of detected features is relevant. However, the capabilities of continuous kernels to process irregularly sampled data are still desired. This article introduces a novel method utilizing continuous kernels for detecting global features at absolute positions in the data domain. Through a use case in processing multiple spatially resolved reflection spectroscopy data, which is sampled irregularly and inconsistently, we show that the proposed method is capable of processing such data natively without additional preprocessing as is needed using comparable methods. In addition, we show that the proposed method is able to achieve a higher prediction accuracy than a comparable network on a dataset with position-dependent features. Furthermore, a higher robustness to missing data compared to a benchmark network using data interpolation is observed, which allows the network to adapt to sensors with individual failed components without the need for retraining.
@inproceedings{magnussen2023utilizing, abstract = {Continuous Kernels have been a recent development in convolutional neural networks. Such kernels are used to process data sampled at different resolutions as well as irregularly and inconsistently sampled data. Convolutional neural networks have the property of translational invariance (e.g., features are detected regardless of their position in the measurement domain), which is unsuitable for certain types of data, where the position of detected features is relevant. However, the capabilities of continuous kernels to process irregularly sampled data are still desired. This article introduces a novel method utilizing continuous kernels for detecting global features at absolute positions in the data domain. Through a use case in processing multiple spatially resolved reflection spectroscopy data, which is sampled irregularly and inconsistently, we show that the proposed method is capable of processing such data natively without additional preprocessing as is needed using comparable methods. In addition, we show that the proposed method is able to achieve a higher prediction accuracy than a comparable network on a dataset with position-dependent features. Furthermore, a higher robustness to missing data compared to a benchmark network using data interpolation is observed, which allows the network to adapt to sensors with individual failed components without the need for retraining.}, author = {Magnussen, Birk Martin and Stern, Claudius and Sick, Bernhard}, booktitle = {International Conference on Autonomic and Autonomous Systems (ICAS)}, keywords = {ReflectionSpectroscopy}, pages = {49–53}, publisher = {ThinkMind}, title = {Utilizing Continuous Kernels for Processing Irregularly and Inconsistently Sampled Data With Position-Dependent Features}, year = 2023 }
Spatio-temporal LPV model of 2D workpiece temperature for Direct Laser Deposition. Pereira, Guilherme; Jelicic, Goran; Kroll, Andreas. In 22nd IFAC World Congress. IFAC, Yokohama, 2023.
- [ BibTeX ]
@inproceedings{Pereira-IFAC-2023, address = {Yokohama}, author = {Pereira, Guilherme and Jelicic, Goran and Kroll, Andreas}, booktitle = {22nd IFAC World Congress}, keywords = {isac-www}, month = {Juli}, note = {accepted}, organization = {IFAC}, title = {Spatio-temporal LPV model of 2D workpiece temperature for Direct Laser Deposition}, year = 2023 }
An Airborne Measurement System to Detect, Locate and Quantify Methane Emissions. Kistner, Lars; Schmoll, Robert; Kroll, Andreas. In Sensor and Measurement Science International Conference (SMSI) 2023. AMA Verband für Sensorik und Messtechnik e.V., Nürnberg, 2023.
- [ BibTeX ]
@inproceedings{Kistner2023, address = {Nürnberg}, author = {Kistner, Lars and Schmoll, Robert and Kroll, Andreas}, booktitle = {Sensor and Measurement Science International Conference (SMSI) 2023}, keywords = {isac-www}, month = {May}, note = {accepted}, organization = {AMA Verband für Sensorik und Messtechnik e.V.}, title = {An Airborne Measurement System to Detect, Locate and Quantify Methane Emissions}, year = 2023 }
Integration of a High-Precision 3D Sensor into a 3D Thermography System. Mendez, Miguel; Schramm, Sebastian; Schmoll, Robert; Kroll, Andreas. In Sensor and Measurement Science International Conference (SMSI) 2023. AMA Verband für Sensorik und Messtechnik e.V., Nürnberg, 2023.
- [ BibTeX ]
@inproceedings{Mendez2023, address = {Nürnberg}, author = {Mendez, Miguel and Schramm, Sebastian and Schmoll, Robert and Kroll, Andreas}, booktitle = {Sensor and Measurement Science International Conference (SMSI) 2023}, keywords = {isac-www}, month = {May}, note = {accepted}, organization = {AMA Verband für Sensorik und Messtechnik e.V.}, title = {Integration of a High-Precision 3D Sensor into a 3D Thermography System}, year = 2023 }
A Digital Twin for Part QualityPrediction and Control in Plastic Injection Molding. Rehmer, Alexander; Klute, Marco; Kroll, Andreas; Heim, Hans-Peter. In Modeling, Identification, and Control for Cyber-Physical Systems Towards Industry 4.0, 1st , P. Mercorelli, W. Zhang, H. Nemati, Y. Zhang (reds.). Elsevier, 2023.
- [ BibTeX ]
- [ URL ]
@inbook{EmergMeth_Rehmer_2022, author = {Rehmer, Alexander and Klute, Marco and Kroll, Andreas and Heim, Hans-Peter}, booktitle = {Modeling, Identification, and Control for Cyber-Physical Systems Towards Industry 4.0}, edition = {1st}, editor = {Mercorelli, Paolo and Zhang, Weicun and Nemati, Hamidreza and Zhang, YuMing}, keywords = {isac-www}, note = {accepted}, publisher = {Elsevier}, series = {Emerging Methodologies and Applications in Modelling, Identification and Control}, title = {A Digital Twin for Part QualityPrediction and Control in Plastic Injection Molding}, year = 2023 }
Model-Based Optimization of Vaccination Strategies in Different Phases of Pandemic Virus Spread. Liu, Z.; Omayrat, M.; Stursberg, O. In Informatics in Control, Automation and Robotics. ICINCO 2021. Lecture Notes in Electrical Engineering, 1006, bll 185–208. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This paper aims at applying optimal control principles to investigate optimal vaccination strategies in different phases of a pandemic. Background of the study is that many countries have started their vaccination procedures against the COVID-19 disease in the beginning of 2021, but supply shortages for the vaccines prevented that everyone could be vaccinated immediately. At the beginning of 2022, in contrast, the vaccine supply was ample, but the effectiveness of different existing vaccines to avoid infection by new virus variants was in doubt, as well as the acceptance of booster doses decreased over time. To account for these effects, two formulations of optimization tasks based on different epidemic models are proposed in this paper. The solution of these tasks determines optimal distribution strategies for available vaccines, and optimized vaccination schemes to reduce the need of booster doses for later phase. Effectiveness of these strategies compared with other popular strategies (as applied in practice) is demonstrated through a series of simulations.
@article{liu2023modelbased, abstract = {This paper aims at applying optimal control principles to investigate optimal vaccination strategies in different phases of a pandemic. Background of the study is that many countries have started their vaccination procedures against the COVID-19 disease in the beginning of 2021, but supply shortages for the vaccines prevented that everyone could be vaccinated immediately. At the beginning of 2022, in contrast, the vaccine supply was ample, but the effectiveness of different existing vaccines to avoid infection by new virus variants was in doubt, as well as the acceptance of booster doses decreased over time. To account for these effects, two formulations of optimization tasks based on different epidemic models are proposed in this paper. The solution of these tasks determines optimal distribution strategies for available vaccines, and optimized vaccination schemes to reduce the need of booster doses for later phase. Effectiveness of these strategies compared with other popular strategies (as applied in practice) is demonstrated through a series of simulations.}, author = {Liu, Z. and Omayrat, M. and Stursberg, O.}, journal = {Informatics in Control, Automation and Robotics. ICINCO 2021. Lecture Notes in Electrical Engineering}, keywords = {isac-www}, pages = {185–208}, title = {Model-Based Optimization of Vaccination Strategies in Different Phases of Pandemic Virus Spread}, volume = 1006, year = 2023 }
Who knows best? A Case Study on Intelligent Crowdworker Selection via Deep Learning. Herde, Marek; Huseljic, Denis; Sick, Bernhard; Bretschneider, Ulrich; Oeste-Reiß, Sarah. In Workshop on Interactive Adapative Learning (IAL), ECML PKDD, bll 14–18. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Crowdworking is a popular approach for annotating large amounts of data to train deep neural networks. However, parts of the annotations are often erroneous. In a case study, we demonstrate how an intelligent crowdworker selection via deep learning reduces the number of erroneous annotations and, thus, the annotation costs of obtaining reliable data for training deep neural networks.
@inproceedings{herde2023who, abstract = {Crowdworking is a popular approach for annotating large amounts of data to train deep neural networks. However, parts of the annotations are often erroneous. In a case study, we demonstrate how an intelligent crowdworker selection via deep learning reduces the number of erroneous annotations and, thus, the annotation costs of obtaining reliable data for training deep neural networks.}, author = {Herde, Marek and Huseljic, Denis and Sick, Bernhard and Bretschneider, Ulrich and Oeste-Reiß, Sarah}, booktitle = {Workshop on Interactive Adapative Learning (IAL), ECML PKDD}, keywords = {ActiveLearning}, pages = {14–18}, title = {Who knows best? A Case Study on Intelligent Crowdworker Selection via Deep Learning}, year = 2023 }
Role of Hyperparameters in Deep Active Learning. Huseljic, Denis; Herde, Marek; Hahn, Paul; Sick, Bernhard. In Workshop on Interactive Adaptive Learning (IAL), ECML PKDD, bll 19–24. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Despite considerable research on pool-based deep active learning (DAL), deploying DAL into real applications still has several challenges. A frequently neglected aspect is the choice of training hyperparameters (HPs), such as the learning rate. Since these HPs determine how the deep neural network learns in each cycle, they must be chosen carefully. In this article, we analyze the role of HPs in DAL. We find that optimizing HPs reduces the performance gap between DAL strategies. Conversely, we highlight challenges in finding optimal HPs when using datasets selected via DAL strategies.
@inproceedings{huseljic2023role, abstract = {Despite considerable research on pool-based deep active learning (DAL), deploying DAL into real applications still has several challenges. A frequently neglected aspect is the choice of training hyperparameters (HPs), such as the learning rate. Since these HPs determine how the deep neural network learns in each cycle, they must be chosen carefully. In this article, we analyze the role of HPs in DAL. We find that optimizing HPs reduces the performance gap between DAL strategies. Conversely, we highlight challenges in finding optimal HPs when using datasets selected via DAL strategies.}, author = {Huseljic, Denis and Herde, Marek and Hahn, Paul and Sick, Bernhard}, booktitle = {Workshop on Interactive Adaptive Learning (IAL), ECML PKDD}, keywords = {isac-www}, pages = {19–24}, title = {Role of Hyperparameters in Deep Active Learning}, year = 2023 }
Spatio-Temporal Attention Graph Neural Network for Remaining Useful Life Prediction. Huang, Zhixin; He, Yujiang; Sick, Bernhard. In Computational Science and Computational Intelligence (CSCI), bll 99–105. IEEE, 2023.
- [ Abstract ]
- [ BibTeX ]
RUL prediction plays a crucial role in the health management of industrial systems. Given the increasing complexity of systems, data-driven predictive models have attracted significant research interest. Upon reviewing the existing literature, it appears that many studies either do not fully integrate both spatial and temporal features or employ only a single attention mechanism. Furthermore, there seems to be inconsistency in the choice of data normalization methods, particularly concerning operating conditions, which might influence predictive performance. To bridge these observations, this study presents the Spatio-Temporal Attention Graph Neural Network. Our model combines graph neural networks and temporal convolutional neural networks for spatial and temporal feature extraction, respectively. The cascade of these extractors, combined with multihead attention mechanisms for both spatio-temporal dimensions, aims to improve predictive precision and refine model explainability. Comprehensive experiments were conducted on the CMAPSS dataset to evaluate the impact of unified versus clustering normalization. The findings suggest that our model performs state-of-the-art results using only the unified normalization. Additionally, when dealing with datasets with multiple operating conditions, cluster normalization enhances the performance of our proposed model by up to 27%.
@inproceedings{huang2023spatio, abstract = {RUL prediction plays a crucial role in the health management of industrial systems. Given the increasing complexity of systems, data-driven predictive models have attracted significant research interest. Upon reviewing the existing literature, it appears that many studies either do not fully integrate both spatial and temporal features or employ only a single attention mechanism. Furthermore, there seems to be inconsistency in the choice of data normalization methods, particularly concerning operating conditions, which might influence predictive performance. To bridge these observations, this study presents the Spatio-Temporal Attention Graph Neural Network. Our model combines graph neural networks and temporal convolutional neural networks for spatial and temporal feature extraction, respectively. The cascade of these extractors, combined with multihead attention mechanisms for both spatio-temporal dimensions, aims to improve predictive precision and refine model explainability. Comprehensive experiments were conducted on the CMAPSS dataset to evaluate the impact of unified versus clustering normalization. The findings suggest that our model performs state-of-the-art results using only the unified normalization. Additionally, when dealing with datasets with multiple operating conditions, cluster normalization enhances the performance of our proposed model by up to 27%.}, author = {Huang, Zhixin and He, Yujiang and Sick, Bernhard}, booktitle = {Computational Science and Computational Intelligence (CSCI)}, keywords = {Clustering_Normalization}, pages = {99–105}, publisher = {IEEE}, title = {Spatio-Temporal Attention Graph Neural Network for Remaining Useful Life Prediction}, year = 2023 }
Time-aware Robustness of Temporal Graph Neural Networks for Link Prediction. Sälzer, Marco; Beddar-Wiesing, Silvia. In International Symposium on Temporal Representation and Reasoning (TIME), bll 19:1–19:3. 2023.
- [ Abstract ]
- [ BibTeX ]
Graph Neural Networks (GNNs) provide a framework for computing functions over graphs based on learnable parameters, which gained much attention in recent years. The most popular GNN models, so-called convolutional GNN or message-passing GNN apply a neighborhood aggregation procedure to each node in a graph to compute its output. Usually, such GNNs are used for classification or prediction tasks over static graphs. However, this limits their applicability in contexts like social networks or knowledge graphs, where underlying graphs change stepwise or time-continuously. Temporal Graph Neural Networks (TGNN) try to close this gap. The general idea of TGNN is to generalize the neighborhood aggregation procedure mentioned above to temporal graphs, usually represented as a tuple of a base graph with a series of time-stamped observed changes. In most applications involving Neural Network based models, giving reliable safety certificates are highly desirable but also a significant challenge, especially because of the blackbox nature of neural models. In this extended abstract, we address the topic of verifying TGNN, which is an unexplored area of research. We present a first notion of a time-aware robustness property for TGNN used for link prediction tasks, motivated by recent work on similar time-aware attacks. Furthermore, we discuss our ongoing work regarding promising verification approaches for the presented or similar safety properties and possible next steps in this research direction.
@inproceedings{salzer2023time, abstract = {Graph Neural Networks (GNNs) provide a framework for computing functions over graphs based on learnable parameters, which gained much attention in recent years. The most popular GNN models, so-called convolutional GNN or message-passing GNN apply a neighborhood aggregation procedure to each node in a graph to compute its output. Usually, such GNNs are used for classification or prediction tasks over static graphs. However, this limits their applicability in contexts like social networks or knowledge graphs, where underlying graphs change stepwise or time-continuously. Temporal Graph Neural Networks (TGNN) try to close this gap. The general idea of TGNN is to generalize the neighborhood aggregation procedure mentioned above to temporal graphs, usually represented as a tuple of a base graph with a series of time-stamped observed changes. In most applications involving Neural Network based models, giving reliable safety certificates are highly desirable but also a significant challenge, especially because of the blackbox nature of neural models. In this extended abstract, we address the topic of verifying TGNN, which is an unexplored area of research. We present a first notion of a time-aware robustness property for TGNN used for link prediction tasks, motivated by recent work on similar time-aware attacks. Furthermore, we discuss our ongoing work regarding promising verification approaches for the presented or similar safety properties and possible next steps in this research direction.}, author = {Sälzer, Marco and Beddar-Wiesing, Silvia}, booktitle = {International Symposium on Temporal Representation and Reasoning (TIME)}, keywords = {isac-www}, pages = {19:1–19:3}, title = {Time-aware Robustness of Temporal Graph Neural Networks for Link Prediction}, year = 2023 }
Organic Computing -- Doctoral Dissertation Colloquium 2022. Tomforde, Sven; Krupitzer, Christian. In Vol. 24Intelligent Embedded Systems. kassel university press, 2023.
- [ BibTeX ]
@book{tomforde2023organic, editor = {Tomforde, Sven and Krupitzer, Christian}, keywords = {isac-www}, publisher = {kassel university press}, series = {Intelligent Embedded Systems}, title = {Organic Computing – Doctoral Dissertation Colloquium 2022}, volume = 24, year = 2023 }
An Examination of Organic Computing Strategies in Design Optimization. Decke, Jens. In Organic Computing -- Doctoral Dissertation Colloquium 2023, S. Tomforde, C. Krupitzer (reds.), bll 41–54. kassel university press, 2023.
- [ Abstract ]
- [ BibTeX ]
In the field of design optimization, numerical equation solvers, commonly used for finite element analysis and computational fluid dy- namics, impose significant computational demands. The iterative nature of design optimization, involving numerous simulations, magnifies re- source requirements in terms of time, energy, costs, and greenhouse gas emissions. Accelerating this process has been a long-standing research challenge, driven by the potential for resource savings and the ability to tackle increasingly complex problems. Recently, organic computing methods have gained prominence as promising approaches to address this challenge. This study aims to bridge the gap by integrating tech- niques from deep learning, active learning, and generative learning into the field of design optimization. The goal is to accelerate the design op- timization process while addressing specific challenges such as exploring large and complex design spaces and evaluating the advantages and con- straints of these methods. This research has the potential to significantly impact the industrial use of design optimization by providing faster and more efficient tools.
@incollection{decke2023examination, abstract = {In the field of design optimization, numerical equation solvers, commonly used for finite element analysis and computational fluid dy- namics, impose significant computational demands. The iterative nature of design optimization, involving numerous simulations, magnifies re- source requirements in terms of time, energy, costs, and greenhouse gas emissions. Accelerating this process has been a long-standing research challenge, driven by the potential for resource savings and the ability to tackle increasingly complex problems. Recently, organic computing methods have gained prominence as promising approaches to address this challenge. This study aims to bridge the gap by integrating tech- niques from deep learning, active learning, and generative learning into the field of design optimization. The goal is to accelerate the design op- timization process while addressing specific challenges such as exploring large and complex design spaces and evaluating the advantages and con- straints of these methods. This research has the potential to significantly impact the industrial use of design optimization by providing faster and more efficient tools.}, author = {Decke, Jens}, booktitle = {Organic Computing – Doctoral Dissertation Colloquium 2023}, editor = {Tomforde, Sven and Krupitzer, Christian}, keywords = {GenerativeLearning}, pages = {41–54}, publisher = {kassel university press}, title = {An Examination of Organic Computing Strategies in Design Optimization}, year = 2023 }
Machine Learning Supported Optimisation and Experimental Evaluation of Electrical Motors for Small Urban Passenger Vehicles. Botache, Diego. In Organic Computing -- Doctoral Dissertation Colloquium 2023, S. Tomforde, C. Krupitzer (reds.), bll 14–26. kassel university press, 2023.
- [ Abstract ]
- [ BibTeX ]
The increasing environmental pollution and rising global energy demands require the rapid development of novel and sustainable traction machines from urban transport up to large-scale transport vehicles. The optimisation of the development cycle of electrical traction machines plays a vital role in sustainable mobility. Still, developers have to face the difficulty of analysing time-consuming multiphysics simulations to assess the performance of many prototype variants. Accelerating this process involves implementing machine learning-supported strategies that can effectively acquire results and predictions about the performance and efficiency of prototypes in an iterative optimisation process and detect possible issues during the experimental procedure. The following research project streamlines the development cycle by identifying and presenting multiple strategies for applying and evaluating machine learning and deep learning techniques in different stages and areas in the development cycle including self-optimisation and self-adaptation capabilities. By implementing these strategies, the optimisation process can be further improved and become more efficient and robust, reducing the time and cost required for testing and validation. Therefore, this proposal addresses two application areas. The first area consists of the supported machine learning motor performance prediction at the topology design and optimisation stage of structural components. The second area corresponds to the machine learning-supported experimental evaluation process of prototypes at desired test benches.
@incollection{botache2023machine, abstract = {The increasing environmental pollution and rising global energy demands require the rapid development of novel and sustainable traction machines from urban transport up to large-scale transport vehicles. The optimisation of the development cycle of electrical traction machines plays a vital role in sustainable mobility. Still, developers have to face the difficulty of analysing time-consuming multiphysics simulations to assess the performance of many prototype variants. Accelerating this process involves implementing machine learning-supported strategies that can effectively acquire results and predictions about the performance and efficiency of prototypes in an iterative optimisation process and detect possible issues during the experimental procedure. The following research project streamlines the development cycle by identifying and presenting multiple strategies for applying and evaluating machine learning and deep learning techniques in different stages and areas in the development cycle including self-optimisation and self-adaptation capabilities. By implementing these strategies, the optimisation process can be further improved and become more efficient and robust, reducing the time and cost required for testing and validation. Therefore, this proposal addresses two application areas. The first area consists of the supported machine learning motor performance prediction at the topology design and optimisation stage of structural components. The second area corresponds to the machine learning-supported experimental evaluation process of prototypes at desired test benches.}, author = {Botache, Diego}, booktitle = {Organic Computing – Doctoral Dissertation Colloquium 2023}, editor = {Tomforde, Sven and Krupitzer, Christian}, keywords = {test_benches}, pages = {14–26}, publisher = {kassel university press}, title = {Machine Learning Supported Optimisation and Experimental Evaluation of Electrical Motors for Small Urban Passenger Vehicles}, year = 2023 }
DADO – Low-Cost Query Strategies for Deep Active Design Optimization. Decke, Jens; Gruhl, Christian; Rauch, Lukas; Sick, Bernhard. In International Conference on Machine Learning and Applications (ICMLA), bll 1611–1618. IEEE, 2023.
- [ Abstract ]
- [ BibTeX ]
In this work, we apply deep active learning to the field of design optimization to reduce the number of computationally expensive numerical simulations widely used in industry and engineering. We are interested in optimizing the design of structural components, where a set of parameters describes the shape. If we can predict the performance based on these parameters and consider only the promising candidates for simulation, there is an enormous potential for saving computing power. We present two query strategies for self-optimization to reduce the computational cost in multi-objective design optimization problems. Our proposed methodology provides an intuitive approach that is easy to apply, offers significant improvements over random sampling, and circumvents the need for uncertainty estimation. We evaluate our strategies on a large dataset from the domain of fluid dynamics and introduce two new evaluation metrics to determine the model's performance. Findings from our evaluation highlights the effectiveness of our query strategies in accelerating design optimization. Furthermore, the introduced method is easily transferable to other self-optimization problems in industry and engineering.
@inproceedings{decke2023dado, abstract = {In this work, we apply deep active learning to the field of design optimization to reduce the number of computationally expensive numerical simulations widely used in industry and engineering. We are interested in optimizing the design of structural components, where a set of parameters describes the shape. If we can predict the performance based on these parameters and consider only the promising candidates for simulation, there is an enormous potential for saving computing power. We present two query strategies for self-optimization to reduce the computational cost in multi-objective design optimization problems. Our proposed methodology provides an intuitive approach that is easy to apply, offers significant improvements over random sampling, and circumvents the need for uncertainty estimation. We evaluate our strategies on a large dataset from the domain of fluid dynamics and introduce two new evaluation metrics to determine the model's performance. Findings from our evaluation highlights the effectiveness of our query strategies in accelerating design optimization. Furthermore, the introduced method is easily transferable to other self-optimization problems in industry and engineering.}, author = {Decke, Jens and Gruhl, Christian and Rauch, Lukas and Sick, Bernhard}, booktitle = {International Conference on Machine Learning and Applications (ICMLA)}, keywords = {Numerical-Simulation}, pages = {1611–1618}, publisher = {IEEE}, title = {DADO – Low-Cost Query Strategies for Deep Active Design Optimization}, year = 2023 }
Towards Few-Shot Time Series Anomaly Detection with Temporal Attention and Dynamic Thresholding. Nivarthi, Chandana Priya; Sick, Bernhard. In International Conference on Machine Learning and Applications (ICMLA), bll 1444–1450. IEEE, 2023.
- [ Abstract ]
- [ BibTeX ]
Anomaly detection plays a pivotal role in diverse realworld applications such as cybersecurity, fault detection, network monitoring, predictive maintenance, and highly automated driving. However, obtaining labeled anomalous data can be a formidable challenge, especially when anomalies exhibit temporal evolution. This paper introduces LATAM (Long short-term memory Autoencoder with Temporal Attention Mechanism) for few-shot anomaly detection, with the aim of enhancing detection performance in scenarios with limited labeled anomaly data. LATAM effectively captures temporal dependencies and emphasizes significant patterns in multivariate time series data. In our investigation, we comprehensively evaluate LATAM against other anomaly detection models, particularly assessing its capability in few-shot learning scenarios where we have minimal examples from the normal class and none from the anomalous class in the training data. Our experimental results, derived from real-world photovoltaic inverter data, highlight LATAM’s superiority, showcasing a substantial 27% mean F1 score improvement, even when trained on a mere two-week dataset. Furthermore, LATAM demonstrates remarkable results on the open-source SWaT dataset, achieving a 12% boost in accuracy with only two days of training data. Moreover, we introduce a simple yet effective dynamic thresholding mechanism, further enhancing the anomaly detection capabilities of LATAM. This underscores LATAM’s efficacy in addressing the challenges posed by limited labeled anomalies in practical scenarios and it proves valuable for downstream tasks involving temporal representation and time series prediction, extending its utility beyond anomaly detection applications.
@inproceedings{nivarthi2023towards, abstract = {Anomaly detection plays a pivotal role in diverse realworld applications such as cybersecurity, fault detection, network monitoring, predictive maintenance, and highly automated driving. However, obtaining labeled anomalous data can be a formidable challenge, especially when anomalies exhibit temporal evolution. This paper introduces LATAM (Long short-term memory Autoencoder with Temporal Attention Mechanism) for few-shot anomaly detection, with the aim of enhancing detection performance in scenarios with limited labeled anomaly data. LATAM effectively captures temporal dependencies and emphasizes significant patterns in multivariate time series data. In our investigation, we comprehensively evaluate LATAM against other anomaly detection models, particularly assessing its capability in few-shot learning scenarios where we have minimal examples from the normal class and none from the anomalous class in the training data. Our experimental results, derived from real-world photovoltaic inverter data, highlight LATAM’s superiority, showcasing a substantial 27% mean F1 score improvement, even when trained on a mere two-week dataset. Furthermore, LATAM demonstrates remarkable results on the open-source SWaT dataset, achieving a 12% boost in accuracy with only two days of training data. Moreover, we introduce a simple yet effective dynamic thresholding mechanism, further enhancing the anomaly detection capabilities of LATAM. This underscores LATAM’s efficacy in addressing the challenges posed by limited labeled anomalies in practical scenarios and it proves valuable for downstream tasks involving temporal representation and time series prediction, extending its utility beyond anomaly detection applications.}, author = {Nivarthi, Chandana Priya and Sick, Bernhard}, booktitle = {International Conference on Machine Learning and Applications (ICMLA)}, keywords = {thresholding}, pages = {1444–1450}, publisher = {IEEE}, title = {Towards Few-Shot Time Series Anomaly Detection with Temporal Attention and Dynamic Thresholding}, year = 2023 }
Context Information for Corner Case Detection in Highly Automated Driving. Heidecker, Florian; Susetzky, Tobias; Fuchs, Erich; Sick, Bernhard. In IEEE International Conference on Intelligent Transportation Systems (ITSC), bll 1522–1529. IEEE, 2023.
- [ Abstract ]
- [ BibTeX ]
Context information provided along with a dataset can be very helpful for solving a problem because the additional knowledge is already available and does not need to be extracted. Moreover, the context indicates how diverse a dataset is, i.e., how many samples per context category are available to train and test machine learning (ML) models. In this article, we present context annotations for the BDD100k image dataset. The annotations comprise, for instance, information about daytime, road condition (dry/wet), and dirt on the windshield. Sometimes, no or only little data are available for unique or rare combinations of these context attributes. However, data that matches these context conditions is crucial when discussing corner cases: Firstly, most ML models, e.g., object detectors, are not trained on such data, which leads to the assumption that they will perform poorly in those situations. Secondly, data containing corner cases are required for validating ML models. With this in mind, separate ML models dedicated to context detection are useful for expanding the training set with additional data of special interest, such as corner cases.
@inproceedings{heidecker2023context, abstract = {Context information provided along with a dataset can be very helpful for solving a problem because the additional knowledge is already available and does not need to be extracted. Moreover, the context indicates how diverse a dataset is, i.e., how many samples per context category are available to train and test machine learning (ML) models. In this article, we present context annotations for the BDD100k image dataset. The annotations comprise, for instance, information about daytime, road condition (dry/wet), and dirt on the windshield. Sometimes, no or only little data are available for unique or rare combinations of these context attributes. However, data that matches these context conditions is crucial when discussing corner cases: Firstly, most ML models, e.g., object detectors, are not trained on such data, which leads to the assumption that they will perform poorly in those situations. Secondly, data containing corner cases are required for validating ML models. With this in mind, separate ML models dedicated to context detection are useful for expanding the training set with additional data of special interest, such as corner cases.}, author = {Heidecker, Florian and Susetzky, Tobias and Fuchs, Erich and Sick, Bernhard}, booktitle = {IEEE International Conference on Intelligent Transportation Systems (ITSC)}, keywords = {isac-www}, pages = {1522–1529}, publisher = {IEEE}, title = {Context Information for Corner Case Detection in Highly Automated Driving}, year = 2023 }
Graph Pooling Provably Improves Expressivity. Lachi, Veronica; Moallemy-Oureh, Alice; Roth, Andreas; Welke, Pascal. In Workshop on New Frontiers in Graph Learning, NeurIPS, bll 1–7. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
In the domain of graph neural networks (GNNs), pooling operators are fundamental to reduce the size of the graph by simplifying graph structures and vertex features. Recent advances have shown that well-designed pooling operators, coupled with message-passing layers, can endow hierarchical GNNs with an expressive power regarding the graph isomorphism test that is equal to the Weisfeiler-Leman test. However, the ability of hierarchical GNNs to increase expressive power by utilizing graph coarsening was not yet explored. This results in uncertainties about the benefits of pooling operators and a lack of sufficient properties to guide their design. In this work, we identify conditions for pooling operators to generate WL-distinguishable coarsened graphs from originally WL-indistinguishable but non-isomorphic graphs. Our conditions are versatile and can be tailored to specific tasks and data characteristics, offering a promising avenue for further research.
@inproceedings{lachi2023graph, abstract = {In the domain of graph neural networks (GNNs), pooling operators are fundamental to reduce the size of the graph by simplifying graph structures and vertex features. Recent advances have shown that well-designed pooling operators, coupled with message-passing layers, can endow hierarchical GNNs with an expressive power regarding the graph isomorphism test that is equal to the Weisfeiler-Leman test. However, the ability of hierarchical GNNs to increase expressive power by utilizing graph coarsening was not yet explored. This results in uncertainties about the benefits of pooling operators and a lack of sufficient properties to guide their design. In this work, we identify conditions for pooling operators to generate WL-distinguishable coarsened graphs from originally WL-indistinguishable but non-isomorphic graphs. Our conditions are versatile and can be tailored to specific tasks and data characteristics, offering a promising avenue for further research.}, author = {Lachi, Veronica and Moallemy-Oureh, Alice and Roth, Andreas and Welke, Pascal}, booktitle = {Workshop on New Frontiers in Graph Learning, NeurIPS}, keywords = {isac-www}, pages = {1–7}, title = {Graph Pooling Provably Improves Expressivity}, year = 2023 }
Leveraging Repeated Unlabelled Noisy Measurements to Augment Supervised Learning. Magnussen, Birk Martin; Stern, Claudius; Sick, Bernhard. In International Conference on Computational Intelligence and Intelligent Systems (CIIS), bll 1–6. ACM, 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Often, producing large labelled datasets for supervised machine learning is difficult and expensive. In cases where the expensive part is due to labelling and obtaining ground truth, it is often comparably easy to acquire large datasets containing unlabelled data points. For reproducible measurements, it is possible to record information on multiple data points being from the same reproducible measurement series, which should thus have an equal but unknown ground truth. In this article, we propose a method to incorporate a dataset of such unlabelled data points for which some data points are known to be equal in end-to-end training of otherwise labelled data. We show that, with the example of predicting the carotenoid concentration in human skin from optical multiple spatially resolved reflection spectroscopy data, the proposed method is capable of reducing the required number of labelled data points to achieve the same prediction accuracy for different model architectures. In addition, we show that the proposed method is capable of reducing the negative impact of noisy data when performing a repeated measurement of the same sample.
@inproceedings{magnussen2023leveraging, abstract = {Often, producing large labelled datasets for supervised machine learning is difficult and expensive. In cases where the expensive part is due to labelling and obtaining ground truth, it is often comparably easy to acquire large datasets containing unlabelled data points. For reproducible measurements, it is possible to record information on multiple data points being from the same reproducible measurement series, which should thus have an equal but unknown ground truth. In this article, we propose a method to incorporate a dataset of such unlabelled data points for which some data points are known to be equal in end-to-end training of otherwise labelled data. We show that, with the example of predicting the carotenoid concentration in human skin from optical multiple spatially resolved reflection spectroscopy data, the proposed method is capable of reducing the required number of labelled data points to achieve the same prediction accuracy for different model architectures. In addition, we show that the proposed method is capable of reducing the negative impact of noisy data when performing a repeated measurement of the same sample.}, author = {Magnussen, Birk Martin and Stern, Claudius and Sick, Bernhard}, booktitle = {International Conference on Computational Intelligence and Intelligent Systems (CIIS)}, keywords = {reflection_spectroscopy}, pages = {1–6}, publisher = {ACM}, title = {Leveraging Repeated Unlabelled Noisy Measurements to Augment Supervised Learning}, year = 2023 }
Multi-Task Representation Learning for Renewable-Power Forecasting: A Comparative Analysis of Unified Autoencoder Variants and Task-Embedding Dimensions. Nivarthi, Chandana Priya; Vogt, Stephan; Sick, Bernhard. In Machine Learning and Knowledge Extraction (MAKE), 5(3), bll 1214–1233. MDPI, 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Typically, renewable-power-generation forecasting using machine learning involves creating separate models for each photovoltaic or wind park, known as single-task learning models. However, transfer learning has gained popularity in recent years, as it allows for the transfer of knowledge from source parks to target parks. Nevertheless, determining the most similar source park(s) for transfer learning can be challenging, particularly when the target park has limited or no historical data samples. To address this issue, we propose a multi-task learning architecture that employs a Unified Autoencoder (UAE) to initially learn a common representation of input weather features among tasks and then utilizes a Task-Embedding layer in a Neural Network (TENN) to learn task-specific information. This proposed UAE-TENN architecture can be easily extended to new parks with or without historical data. We evaluate the performance of our proposed architecture and compare it to single-task learning models on six photovoltaic and wind farm datasets consisting of a total of 529 parks. Our results show that the UAE-TENN architecture significantly improves power-forecasting performance by 10 to 19% for photovoltaic parks and 5 to 15% for wind parks compared to baseline models. We also demonstrate that UAE-TENN improves forecast accuracy for a new park by 19% for photovoltaic parks, even in a zero-shot learning scenario where there is no historical data. Additionally, we propose variants of the Unified Autoencoder with convolutional and LSTM layers, compare their performance, and provide a comparison among architectures with different numbers of task-embedding dimensions. Finally, we demonstrate the utility of trained task embeddings for interpretation and visualization purposes.
@article{nivarthi2023multi, abstract = {Typically, renewable-power-generation forecasting using machine learning involves creating separate models for each photovoltaic or wind park, known as single-task learning models. However, transfer learning has gained popularity in recent years, as it allows for the transfer of knowledge from source parks to target parks. Nevertheless, determining the most similar source park(s) for transfer learning can be challenging, particularly when the target park has limited or no historical data samples. To address this issue, we propose a multi-task learning architecture that employs a Unified Autoencoder (UAE) to initially learn a common representation of input weather features among tasks and then utilizes a Task-Embedding layer in a Neural Network (TENN) to learn task-specific information. This proposed UAE-TENN architecture can be easily extended to new parks with or without historical data. We evaluate the performance of our proposed architecture and compare it to single-task learning models on six photovoltaic and wind farm datasets consisting of a total of 529 parks. Our results show that the UAE-TENN architecture significantly improves power-forecasting performance by 10 to 19% for photovoltaic parks and 5 to 15% for wind parks compared to baseline models. We also demonstrate that UAE-TENN improves forecast accuracy for a new park by 19% for photovoltaic parks, even in a zero-shot learning scenario where there is no historical data. Additionally, we propose variants of the Unified Autoencoder with convolutional and LSTM layers, compare their performance, and provide a comparison among architectures with different numbers of task-embedding dimensions. Finally, we demonstrate the utility of trained task embeddings for interpretation and visualization purposes.}, author = {Nivarthi, Chandana Priya and Vogt, Stephan and Sick, Bernhard}, journal = {Machine Learning and Knowledge Extraction (MAKE)}, keywords = {isac-www}, number = 3, pages = {1214–1233}, publisher = {MDPI}, title = {Multi-Task Representation Learning for Renewable-Power Forecasting: A Comparative Analysis of Unified Autoencoder Variants and Task-Embedding Dimensions}, volume = 5, year = 2023 }
Height Change Feature Based Free Space Detection. Schreck, Steven; Reichert, Hannes; Hetzel, Manuel; Doll, Konrad; Sick, Bernhard. In International Conference on Control, Mechatronics and Automation (ICCMA), bll 171–176. IEEE, 2023.
- [ Abstract ]
- [ BibTeX ]
We present a novel method for free space detection in dynamic environments like factory sites, crucial for autonomous forklifts to avoid collisions. It introduces a technique for fast surface normal estimation using spherical projected LiDAR data, which helps in detecting free space efficiently and in real-time. The method's effectiveness is proven with a 50.90% mIoU score on the Semantic KITTI dataset at 105 Hz, and a 63.30% mIoU score on a factory site dataset at 54 Hz.
@inproceedings{schreck2023height, abstract = {We present a novel method for free space detection in dynamic environments like factory sites, crucial for autonomous forklifts to avoid collisions. It introduces a technique for fast surface normal estimation using spherical projected LiDAR data, which helps in detecting free space efficiently and in real-time. The method's effectiveness is proven with a 50.90% mIoU score on the Semantic KITTI dataset at 105 Hz, and a 63.30% mIoU score on a factory site dataset at 54 Hz.}, author = {Schreck, Steven and Reichert, Hannes and Hetzel, Manuel and Doll, Konrad and Sick, Bernhard}, booktitle = {International Conference on Control, Mechatronics and Automation (ICCMA)}, keywords = {isac-www}, pages = {171–176}, publisher = {IEEE}, title = {Height Change Feature Based Free Space Detection}, year = 2023 }
What Does Really Count? Estimating Relevance of Corner Cases for Semantic Segmentation in Automated Driving. Breitenstein, Jasmin; Heidecker, Florian; Lyssenko, Maria; Bogdoll, Daniel; Bieshaar, Maarten; Zöllner, J. Marius; Sick, Bernhard; Fingscheidt, Tim. In Workshop on roBustness and Reliability of Autonomous Vehicles in the Open-world (BRAVO), ICCV, bll 3991–4000. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
In safety-critical applications such as automated driving, perception errors may create an imminent risk to vulnerable road users (VRU). To mitigate the occurrence of unexpected and potentially dangerous situations, so-called corner cases, perception models are trained on a huge amount of data. However, the models are typically evaluated using task-agnostic metrics, which do not reflect the severity of safety-critical misdetections. Consequently, misdetections with particular relevance for the safe driving task should entail a more severe penalty during evaluation to pinpoint corner cases in large-scale datasets. In this work, we propose a novel metric IoUw that exploits relevance on the pixel level of the semantic segmentation output to extend the notion of the intersection over union (IoU) by emphasizing small areas of an image affected by corner cases. We (i) employ IoUw to measure the effect of pre-defined relevance criteria on the segmentation evaluation, and (ii) use the relevance-adapted IoUw to refine the identification of corner cases. In our experiments, we investigate vision-based relevance criteria and physical attributes as per-pixel criticality to factor in the imminent risk, showing that IoUw precisely accentuates the relevance of corner cases.
@inproceedings{breitenstein2023what, abstract = {In safety-critical applications such as automated driving, perception errors may create an imminent risk to vulnerable road users (VRU). To mitigate the occurrence of unexpected and potentially dangerous situations, so-called corner cases, perception models are trained on a huge amount of data. However, the models are typically evaluated using task-agnostic metrics, which do not reflect the severity of safety-critical misdetections. Consequently, misdetections with particular relevance for the safe driving task should entail a more severe penalty during evaluation to pinpoint corner cases in large-scale datasets. In this work, we propose a novel metric IoUw that exploits relevance on the pixel level of the semantic segmentation output to extend the notion of the intersection over union (IoU) by emphasizing small areas of an image affected by corner cases. We (i) employ IoUw to measure the effect of pre-defined relevance criteria on the segmentation evaluation, and (ii) use the relevance-adapted IoUw to refine the identification of corner cases. In our experiments, we investigate vision-based relevance criteria and physical attributes as per-pixel criticality to factor in the imminent risk, showing that IoUw precisely accentuates the relevance of corner cases.}, author = {Breitenstein, Jasmin and Heidecker, Florian and Lyssenko, Maria and Bogdoll, Daniel and Bieshaar, Maarten and Zöllner, J. Marius and Sick, Bernhard and Fingscheidt, Tim}, booktitle = {Workshop on roBustness and Reliability of Autonomous Vehicles in the Open-world (BRAVO), ICCV}, keywords = {isac-www}, pages = {3991–4000}, title = {What Does Really Count? Estimating Relevance of Corner Cases for Semantic Segmentation in Automated Driving}, year = 2023 }
Targeted Adversarial Attacks on Wind Power Forecasts. Heinrich, René; Scholz, Christoph; Vogt, Stephan; Lehna, Malte. In Machine Learning, 113(2), bll 863–889. Springer, 2023.
- [ Abstract ]
- [ BibTeX ]
In recent years, researchers proposed a variety of deep learning models for wind power forecasting. These models predict the wind power generation of wind farms or entire regions more accurately than traditional machine learning algorithms or physical models. However, latest research has shown that deep learning models can often be manipulated by adversarial attacks. Since wind power forecasts are essential for the stability of modern power systems, it is important to protect them from this threat. In this work, we investigate the vulnerability of two different forecasting models to targeted, semi-targeted, and untargeted adversarial attacks. We consider a Long Short-Term Memory (LSTM) network for predicting the power generation of individual wind farms and a Convolutional Neural Network (CNN) for forecasting the wind power generation throughout Germany. Moreover, we propose the Total Adversarial Robustness Score (TARS), an evaluation metric for quantifying the robustness of regression models to targeted and semi-targeted adversarial attacks. It assesses the impact of attacks on the model's performance, as well as the extent to which the attacker's goal was achieved, by assigning a score between 0 (very vulnerable) and 1 (very robust). In our experiments, the LSTM forecasting model was fairly robust and achieved a TARS value of over 0.78 for all adversarial attacks investigated. The CNN forecasting model only achieved TARS values below 0.10 when trained ordinarily, and was thus very vulnerable. Yet, its robustness could be significantly improved by adversarial training, which always resulted in a TARS above 0.46.
@article{heinrich2023targeted, abstract = {In recent years, researchers proposed a variety of deep learning models for wind power forecasting. These models predict the wind power generation of wind farms or entire regions more accurately than traditional machine learning algorithms or physical models. However, latest research has shown that deep learning models can often be manipulated by adversarial attacks. Since wind power forecasts are essential for the stability of modern power systems, it is important to protect them from this threat. In this work, we investigate the vulnerability of two different forecasting models to targeted, semi-targeted, and untargeted adversarial attacks. We consider a Long Short-Term Memory (LSTM) network for predicting the power generation of individual wind farms and a Convolutional Neural Network (CNN) for forecasting the wind power generation throughout Germany. Moreover, we propose the Total Adversarial Robustness Score (TARS), an evaluation metric for quantifying the robustness of regression models to targeted and semi-targeted adversarial attacks. It assesses the impact of attacks on the model's performance, as well as the extent to which the attacker's goal was achieved, by assigning a score between 0 (very vulnerable) and 1 (very robust). In our experiments, the LSTM forecasting model was fairly robust and achieved a TARS value of over 0.78 for all adversarial attacks investigated. The CNN forecasting model only achieved TARS values below 0.10 when trained ordinarily, and was thus very vulnerable. Yet, its robustness could be significantly improved by adversarial training, which always resulted in a TARS above 0.46.}, author = {Heinrich, René and Scholz, Christoph and Vogt, Stephan and Lehna, Malte}, journal = {Machine Learning}, keywords = {isac-www}, number = 2, pages = {863–889}, publisher = {Springer}, title = {Targeted Adversarial Attacks on Wind Power Forecasts}, volume = 113, year = 2023 }
On the Reliability of Automated Analysis of Fracture Surfaces Using a Novel Computer Vision-Based Tool. Engelhardt, Anna; Decke, Jens; Meier, David; Dulig, Franz; Ragunathan, Rishan; Wegener, Thomas; Sick, Bernhard; Niendorf, Thomas. In Advanced Engineering Materials, 25(21), bl 2300876. Wiley, 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Fracture surface analysis is of utmost importance with respect to structural integrity of metallic materials. This especially holds true for additively manufactured materials. Despite an increasing trend of automatization of testing methods, the analysis and classification of fatigue fracture surface images is commonly done manually by experts. Although this leads to correct results in most cases, it has several disadvantages, e.g., the need of a huge knowledge base to interpret images correctly. In present work, an unsupervised tool for analysis of overview images of fatigue fracture surface images is developed to support nonexperienced users to identify the origin of the fracture. The tool is developed using fracture surface images of additively manufactured Ti6Al4V specimens fatigued in the high-cycle-fatigue regime and is based on the identification of river marks. Several recording parameters seem to have no significant influence on the results as long as preprocessing settings are adapted. Moreover, it is possible to analyze images of other materials with the tool as long as the fracture surfaces contain river marks. However, special features like multiple origins or origins located in direct vicinity to the surface, e.g., caused by increased plastic strains, require a further tool development or alternative approaches.
@article{engelhardt2023reliability, abstract = {Fracture surface analysis is of utmost importance with respect to structural integrity of metallic materials. This especially holds true for additively manufactured materials. Despite an increasing trend of automatization of testing methods, the analysis and classification of fatigue fracture surface images is commonly done manually by experts. Although this leads to correct results in most cases, it has several disadvantages, e.g., the need of a huge knowledge base to interpret images correctly. In present work, an unsupervised tool for analysis of overview images of fatigue fracture surface images is developed to support nonexperienced users to identify the origin of the fracture. The tool is developed using fracture surface images of additively manufactured Ti6Al4V specimens fatigued in the high-cycle-fatigue regime and is based on the identification of river marks. Several recording parameters seem to have no significant influence on the results as long as preprocessing settings are adapted. Moreover, it is possible to analyze images of other materials with the tool as long as the fracture surfaces contain river marks. However, special features like multiple origins or origins located in direct vicinity to the surface, e.g., caused by increased plastic strains, require a further tool development or alternative approaches.}, author = {Engelhardt, Anna and Decke, Jens and Meier, David and Dulig, Franz and Ragunathan, Rishan and Wegener, Thomas and Sick, Bernhard and Niendorf, Thomas}, journal = {Advanced Engineering Materials}, keywords = {Ti6Al4V}, number = 21, pages = 2300876, publisher = {Wiley}, title = {On the Reliability of Automated Analysis of Fracture Surfaces Using a Novel Computer Vision-Based Tool}, volume = 25, year = 2023 }
Continuous Feature Networks: A Novel Method to Process Irregularly and Inconsistently Sampled Data With Position-Dependent Features. Magnussen, Birk Martin; Stern, Claudius; Sick, Bernhard. In International Journal On Advances in Intelligent Systems, 16(3&4), bll 43–50. ThinkMind, 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Continuous Kernels have been a recent development in convolutional neural networks. Such kernels are used to process data sampled at different resolutions as well as irregularly and inconsistently sampled data. Convolutional neural networks have the property of translational invariance (e.g., features are detected regardless of their position in the measurement domain), which is unsuitable if the position of detected features is relevant for the prediction task. However, the capabilities of continuous kernels to process irregularly sampled data are still desired. This article introduces the continuous feature network, a novel method utilizing continuous kernels, for detecting global features at absolute positions in the data domain. Through a use case in processing multiple spatially resolved reflection spectroscopy data, which is sampled irregularly and inconsistently, we show that the proposed method is capable of processing such data directly without additional preprocessing or augmentation as is needed using comparable methods. In addition, we show that the proposed method is able to achieve a higher prediction accuracy than a comparable network on a dataset with position-dependent features. Furthermore, a higher robustness to missing data compared to a benchmark network using data interpolation is observed, which allows the network to adapt to sensors with a failure of individual light emitters or detectors without the need for retraining. The article shows how these capabilities stem from the continuous kernels used and how the number of available kernels to be trained affects the model. Finally, the article proposes a method to utilize the introduced method as a base for an interpretable model usable for explainable AI.
@article{magnussen2023continuous, abstract = {Continuous Kernels have been a recent development in convolutional neural networks. Such kernels are used to process data sampled at different resolutions as well as irregularly and inconsistently sampled data. Convolutional neural networks have the property of translational invariance (e.g., features are detected regardless of their position in the measurement domain), which is unsuitable if the position of detected features is relevant for the prediction task. However, the capabilities of continuous kernels to process irregularly sampled data are still desired. This article introduces the continuous feature network, a novel method utilizing continuous kernels, for detecting global features at absolute positions in the data domain. Through a use case in processing multiple spatially resolved reflection spectroscopy data, which is sampled irregularly and inconsistently, we show that the proposed method is capable of processing such data directly without additional preprocessing or augmentation as is needed using comparable methods. In addition, we show that the proposed method is able to achieve a higher prediction accuracy than a comparable network on a dataset with position-dependent features. Furthermore, a higher robustness to missing data compared to a benchmark network using data interpolation is observed, which allows the network to adapt to sensors with a failure of individual light emitters or detectors without the need for retraining. The article shows how these capabilities stem from the continuous kernels used and how the number of available kernels to be trained affects the model. Finally, the article proposes a method to utilize the introduced method as a base for an interpretable model usable for explainable AI.}, author = {Magnussen, Birk Martin and Stern, Claudius and Sick, Bernhard}, journal = {International Journal On Advances in Intelligent Systems}, keywords = {reflection_spectroscopy}, number = {3&4}, pages = {43–50}, publisher = {ThinkMind}, title = {Continuous Feature Networks: A Novel Method to Process Irregularly and Inconsistently Sampled Data With Position-Dependent Features}, volume = 16, year = 2023 }
Corner Cases in Machine Learning Processes. Heidecker, Florian; Bieshaar, Maarten; Sick, Bernhard. In AI Perspectives & Advances, 6(1), bll 1–17. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Applications using machine learning (ML), such as highly autonomous driving, depend highly on the performance of the ML model. The data amount and quality used for model training and validation are crucial. If the model cannot detect and interpret a new, rare, or perhaps dangerous situation, often referred to as a corner case, we will likely blame the data for not being good enough or too small in number. However, the implemented ML model and its associated architecture also influence the behavior. Therefore, the occurrence of prediction errors resulting from the ML model itself is not surprising. This work addresses a corner case definition from an ML model's perspective to determine which aspects must be considered. To achieve this goal, we present an overview of properties for corner cases that are beneficial for the description, explanation, reproduction, or synthetic generation of corner cases. To define ML corner cases, we review different considerations in the literature and summarize them in a general description and mathematical formulation, whereby the expected relevance-weighted loss is the key to distinguishing corner cases from common data. Moreover, we show how to operationalize the corner case characteristics to determine the value of a corner case. To conclude, we present the extended taxonomy for ML corner cases by adding the input, model, and deployment levels, considering the influence of the corner case properties.
@article{heidecker2023corner, abstract = {Applications using machine learning (ML), such as highly autonomous driving, depend highly on the performance of the ML model. The data amount and quality used for model training and validation are crucial. If the model cannot detect and interpret a new, rare, or perhaps dangerous situation, often referred to as a corner case, we will likely blame the data for not being good enough or too small in number. However, the implemented ML model and its associated architecture also influence the behavior. Therefore, the occurrence of prediction errors resulting from the ML model itself is not surprising. This work addresses a corner case definition from an ML model's perspective to determine which aspects must be considered. To achieve this goal, we present an overview of properties for corner cases that are beneficial for the description, explanation, reproduction, or synthetic generation of corner cases. To define ML corner cases, we review different considerations in the literature and summarize them in a general description and mathematical formulation, whereby the expected relevance-weighted loss is the key to distinguishing corner cases from common data. Moreover, we show how to operationalize the corner case characteristics to determine the value of a corner case. To conclude, we present the extended taxonomy for ML corner cases by adding the input, model, and deployment levels, considering the influence of the corner case properties.}, author = {Heidecker, Florian and Bieshaar, Maarten and Sick, Bernhard}, journal = {AI Perspectives & Advances}, keywords = {Taxonomy}, number = 1, pages = {1–17}, title = {Corner Cases in Machine Learning Processes}, volume = 6, year = 2023 }
Towards Enhancing Deep Active Learning with Weak Supervision and Constrained Clustering. Aßenmacher, Matthias; Rauch, Lukas; Goschenhofer, Jann; Stephan, Andreas; Bischl, Bernd; Roth, Benjamin; Sick, Bernhard. In Workshop on Interactive Adapative Learning (IAL), ECML PKDD, bll 65–73. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Three fields revolving around the question of how to cope with limited amounts of labeled data are Deep Active Learning (DAL), deep Constrained Clustering (CC), and Weakly Supervised Learning (WSL). DAL tackles the problem by adaptively posing the question of which data samples to annotate next in order to achieve the best incremental learning improvement, although it suffers from several limitations that hinder its deployment in practical settings. We point out how CC algorithms and WSL could be employed to overcome these limitations and increase the practical applicability of DAL research. Specifically, we discuss the opportunities to use the class discovery capabilities of CC and the possibility of further reducing human annotation efforts by utilizing WSL. We argue that the practical applicability of DAL algorithms will benefit from employing CC and WSL methods for the learning and labeling process. We inspect the overlaps between the three research areas and identify relevant and exciting research questions at the intersection of these areas.
@inproceedings{assenmacher2023towards, abstract = {Three fields revolving around the question of how to cope with limited amounts of labeled data are Deep Active Learning (DAL), deep Constrained Clustering (CC), and Weakly Supervised Learning (WSL). DAL tackles the problem by adaptively posing the question of which data samples to annotate next in order to achieve the best incremental learning improvement, although it suffers from several limitations that hinder its deployment in practical settings. We point out how CC algorithms and WSL could be employed to overcome these limitations and increase the practical applicability of DAL research. Specifically, we discuss the opportunities to use the class discovery capabilities of CC and the possibility of further reducing human annotation efforts by utilizing WSL. We argue that the practical applicability of DAL algorithms will benefit from employing CC and WSL methods for the learning and labeling process. We inspect the overlaps between the three research areas and identify relevant and exciting research questions at the intersection of these areas.}, author = {Aßenmacher, Matthias and Rauch, Lukas and Goschenhofer, Jann and Stephan, Andreas and Bischl, Bernd and Roth, Benjamin and Sick, Bernhard}, booktitle = {Workshop on Interactive Adapative Learning (IAL), ECML PKDD}, keywords = {isac-www}, pages = {65–73}, title = {Towards Enhancing Deep Active Learning with Weak Supervision and Constrained Clustering}, year = 2023 }
It is the habit not the handle that affects tooth brushing. Results of a randomised counterbalanced cross over study. Deinzer, Renate; Eidenhardt, Zdenka; Sohrabi, Keywan; Stenger, Manuel; Kraft, Dominik; Sick, Bernhard; Götz-Hahn, Franz; Bottenbruch, Carlotta; Berneburg, Nils; Weik, Ulrike. In Research Square. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Objective: To assess the effect of the toothbrush handle on video-observed toothbrushing behaviour and toothbrushing effectiveness Methods: This is a randomized counterbalanced cross-over study. N = 50 university students brushed their teeth at two occasions, one week apart, using either a commercial ergonomically designed manual toothbrush (MT) or Brushalyze V1 (BV1), a manual toothbrush with a thick cylindrical handle without any specific ergonomic features. Brushing behaviour was video-analysed. Plaque was assessed at the second occasion immediately after brushing. Participants directly compared the two brushes regarding their handling and compared them to the brushed they used at home. Results: The study participants found the BV1 significantly more cumbersome than the M1 or their brush at home. (p < 0.05). However, correlation analyses revealed a strong consistency of brushing behavior with the two brushes (0.71 < r < 0.91). Means differed only slightly (all d < 0.36). These differences became statistically significant only for the brushing time at inner surfaces (d = 0.31 p = 0.03) and horizontal movements at inner surfaces (d = 0.35, p = 0.02). Plaque levels at the gingival margins did not differ while slightly more plaque persisted at the more coronal aspects of the crown after brushing with BV1 (d = 0.592; p 0.042) Discussion: The results of the study indicate that the brushing handle does not play a major role in brushing behavior or brushing effectiveness.
@article{deinzer2023it, abstract = {Objective: To assess the effect of the toothbrush handle on video-observed toothbrushing behaviour and toothbrushing effectiveness Methods: This is a randomized counterbalanced cross-over study. N = 50 university students brushed their teeth at two occasions, one week apart, using either a commercial ergonomically designed manual toothbrush (MT) or Brushalyze V1 (BV1), a manual toothbrush with a thick cylindrical handle without any specific ergonomic features. Brushing behaviour was video-analysed. Plaque was assessed at the second occasion immediately after brushing. Participants directly compared the two brushes regarding their handling and compared them to the brushed they used at home. Results: The study participants found the BV1 significantly more cumbersome than the M1 or their brush at home. (p < 0.05). However, correlation analyses revealed a strong consistency of brushing behavior with the two brushes (0.71 < r < 0.91). Means differed only slightly (all d < 0.36). These differences became statistically significant only for the brushing time at inner surfaces (d = 0.31 p = 0.03) and horizontal movements at inner surfaces (d = 0.35, p = 0.02). Plaque levels at the gingival margins did not differ while slightly more plaque persisted at the more coronal aspects of the crown after brushing with BV1 (d = 0.592; p 0.042) Discussion: The results of the study indicate that the brushing handle does not play a major role in brushing behavior or brushing effectiveness.}, author = {Deinzer, Renate and Eidenhardt, Zdenka and Sohrabi, Keywan and Stenger, Manuel and Kraft, Dominik and Sick, Bernhard and Götz-Hahn, Franz and Bottenbruch, Carlotta and Berneburg, Nils and Weik, Ulrike}, journal = {Research Square}, keywords = {isac-www}, note = {(preprint)}, title = {It is the habit not the handle that affects tooth brushing. Results of a randomised counterbalanced cross over study}, year = 2023 }
Model selection, adaptation, and combination for transfer learning in wind and photovoltaic power forecasts. Schreiber, Jens; Sick, Bernhard. In Energy and AI, 14, bl 100249. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
There is recent interest in using model hubs – a collection of pre-trained models – in computer vision tasks. To employ a model hub, we first select a source model and then adapt the model for the target to compensate for differences. There still needs to be more research on model selection and adaption for renewable power forecasts. In particular, none of the related work examines different model selection and adaptation strategies for neural network architectures. Also, none of the current studies investigates the influence of available training samples and considers seasonality in the evaluation. We close these gaps by conducting the first thorough experiment for model selection and adaptation for transfer learning in renewable power forecast, adopting recent developments from the field of computer vision on 667 wind and photovoltaic parks from six datasets. We simulate different amounts of training samples for each season to calculate informative forecast errors. We examine the marginal likelihood and forecast error for model selection for those amounts. Furthermore, we study four adaption strategies. As an extension of the current state of the art, we utilize a Bayesian linear regression for forecasting the response based on features extracted from a neural network. This approach outperforms the baseline with only seven days of training data and shows that fine-tuning is not beneficial with less than three months of data. We further show how combining multiple models through ensembles can significantly improve the model selection and adaptation approach such that we have a similar mean error with only 30 days of training data which is otherwise only possible with an entire year of training data. We achieve a mean error of 9.8 and 14 percent for the most realistic dataset for PV and wind with only seven days of training data.
@article{schreiber2023model, abstract = {There is recent interest in using model hubs – a collection of pre-trained models – in computer vision tasks. To employ a model hub, we first select a source model and then adapt the model for the target to compensate for differences. There still needs to be more research on model selection and adaption for renewable power forecasts. In particular, none of the related work examines different model selection and adaptation strategies for neural network architectures. Also, none of the current studies investigates the influence of available training samples and considers seasonality in the evaluation. We close these gaps by conducting the first thorough experiment for model selection and adaptation for transfer learning in renewable power forecast, adopting recent developments from the field of computer vision on 667 wind and photovoltaic parks from six datasets. We simulate different amounts of training samples for each season to calculate informative forecast errors. We examine the marginal likelihood and forecast error for model selection for those amounts. Furthermore, we study four adaption strategies. As an extension of the current state of the art, we utilize a Bayesian linear regression for forecasting the response based on features extracted from a neural network. This approach outperforms the baseline with only seven days of training data and shows that fine-tuning is not beneficial with less than three months of data. We further show how combining multiple models through ensembles can significantly improve the model selection and adaptation approach such that we have a similar mean error with only 30 days of training data which is otherwise only possible with an entire year of training data. We achieve a mean error of 9.8 and 14 percent for the most realistic dataset for PV and wind with only seven days of training data.}, author = {Schreiber, Jens and Sick, Bernhard}, journal = {Energy and AI}, keywords = {isac-www}, pages = 100249, title = {Model selection, adaptation, and combination for transfer learning in wind and photovoltaic power forecasts}, volume = 14, year = 2023 }
Active Label Refinement for Semantic Segmentation of Satellite Images. Pham, Minh Tuan; Wijesingha, Jayan; Kottke, Daniel; Herde, Marek; Huseljic, Denis; Sick, Bernhard; Wachendorf, Michael; Esch, Thomas. In arXiv e-prints, bl arXiv:2309.06159. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Remote sensing through semantic segmentation of satellite images contributes to the understanding and utilisation of the earth's surface. For this purpose, semantic segmentation networks are typically trained on large sets of labelled satellite images. However, obtaining expert labels for these images is costly. Therefore, we propose to rely on a low-cost approach, e.g. crowdsourcing or pretrained networks, to label the images in the first step. Since these initial labels are partially erroneous, we use active learning strategies to cost-efficiently refine the labels in the second step. We evaluate the active learning strategies using satellite images of Bengaluru in India, labelled with land cover and land use labels. Our experimental results suggest that an active label refinement to improve the semantic segmentation network's performance is beneficial.
@article{pham2023active, abstract = {Remote sensing through semantic segmentation of satellite images contributes to the understanding and utilisation of the earth's surface. For this purpose, semantic segmentation networks are typically trained on large sets of labelled satellite images. However, obtaining expert labels for these images is costly. Therefore, we propose to rely on a low-cost approach, e.g. crowdsourcing or pretrained networks, to label the images in the first step. Since these initial labels are partially erroneous, we use active learning strategies to cost-efficiently refine the labels in the second step. We evaluate the active learning strategies using satellite images of Bengaluru in India, labelled with land cover and land use labels. Our experimental results suggest that an active label refinement to improve the semantic segmentation network's performance is beneficial.}, author = {Pham, Minh Tuan and Wijesingha, Jayan and Kottke, Daniel and Herde, Marek and Huseljic, Denis and Sick, Bernhard and Wachendorf, Michael and Esch, Thomas}, journal = {arXiv e-prints}, keywords = {isac-www}, pages = {arXiv:2309.06159}, title = {Active Label Refinement for Semantic Segmentation of Satellite Images}, year = 2023 }
ActiveGLAE: A Benchmark for Deep Active Learning with Transformers. Rauch, Lukas; Aßenmacher, Matthias; Huseljic, Denis; Wirth, Moritz; Bischl, Bernd; Sick, Bernhard. In European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECMLPKDD), bll 55–74. Springer, 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Deep active learning (DAL) seeks to reduce annotation costs by enabling the model to actively query instance annotations from which it expects to learn the most. Despite extensive research, there is cur- rently no standardized evaluation protocol for transformer-based lan- guage models in the field of DAL. Diverse experimental settings lead to difficulties in comparing research and deriving recommendations for prac- titioners. To tackle this challenge, we propose the ActiveGLAE bench- mark, a comprehensive collection of data sets and evaluation guidelines for assessing DAL. Our benchmark aims to facilitate and streamline the evaluation process of novel DAL strategies. Additionally, we provide an extensive overview of current practice in DAL with transformer-based language models. We identify three key challenges - data set selection, model training, and DAL settings - that pose difficulties in comparing query strategies. We establish baseline results through an extensive set of experiments as a reference point for evaluating future work. Based on our findings, we provide guidelines for researchers and practitioners.
@inproceedings{rauch2023activeglae, abstract = {Deep active learning (DAL) seeks to reduce annotation costs by enabling the model to actively query instance annotations from which it expects to learn the most. Despite extensive research, there is cur- rently no standardized evaluation protocol for transformer-based lan- guage models in the field of DAL. Diverse experimental settings lead to difficulties in comparing research and deriving recommendations for prac- titioners. To tackle this challenge, we propose the ActiveGLAE bench- mark, a comprehensive collection of data sets and evaluation guidelines for assessing DAL. Our benchmark aims to facilitate and streamline the evaluation process of novel DAL strategies. Additionally, we provide an extensive overview of current practice in DAL with transformer-based language models. We identify three key challenges - data set selection, model training, and DAL settings - that pose difficulties in comparing query strategies. We establish baseline results through an extensive set of experiments as a reference point for evaluating future work. Based on our findings, we provide guidelines for researchers and practitioners.}, author = {Rauch, Lukas and Aßenmacher, Matthias and Huseljic, Denis and Wirth, Moritz and Bischl, Bernd and Sick, Bernhard}, booktitle = {European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECMLPKDD)}, keywords = {isac-www}, pages = {55–74}, publisher = {Springer}, title = {ActiveGLAE: A Benchmark for Deep Active Learning with Transformers}, year = 2023 }
Active Bird2Vec: Towards End-To-End Bird Sound Monitoring with Transformers. Rauch, Lukas; Schwinger, Raphael; Wirth, Moritz; Sick, Bernhard; Tomforde, Sven; Scholz, Christoph. In Workshop on Artificial Intelligence for Sustainability (AI4S), ECAI, bll 1–6. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
We propose a shift towards end-to-end learning in bird sound monitoring by combining self-supervised (SSL) and deep active learning (DAL). Leveraging transformer models, we aim to bypass traditional spectrogram conversions, enabling direct raw audio processing. ACTIVE BIRD2VEC is set to generate high-quality bird sound representations through SSL, potentially accelerating the assessment of environmental changes and decision-making processes for wind farms. Additionally, we seek to utilize the wide variety of bird vocalizations through DAL, reducing the reliance on extensively labeled datasets by human experts. We plan to curate a comprehensive set of tasks through Huggingface Datasets, enhancing future comparability and reproducibility of bioacoustic research. A comparative analysis between various transformer models will be conducted to evaluate their proficiency in bird sound recognition tasks. We aim to accelerate the progression of avian bioacoustic research and contribute to more effective conservation strategies.
@inproceedings{rauch2023active, abstract = {We propose a shift towards end-to-end learning in bird sound monitoring by combining self-supervised (SSL) and deep active learning (DAL). Leveraging transformer models, we aim to bypass traditional spectrogram conversions, enabling direct raw audio processing. ACTIVE BIRD2VEC is set to generate high-quality bird sound representations through SSL, potentially accelerating the assessment of environmental changes and decision-making processes for wind farms. Additionally, we seek to utilize the wide variety of bird vocalizations through DAL, reducing the reliance on extensively labeled datasets by human experts. We plan to curate a comprehensive set of tasks through Huggingface Datasets, enhancing future comparability and reproducibility of bioacoustic research. A comparative analysis between various transformer models will be conducted to evaluate their proficiency in bird sound recognition tasks. We aim to accelerate the progression of avian bioacoustic research and contribute to more effective conservation strategies.}, author = {Rauch, Lukas and Schwinger, Raphael and Wirth, Moritz and Sick, Bernhard and Tomforde, Sven and Scholz, Christoph}, booktitle = {Workshop on Artificial Intelligence for Sustainability (AI4S), ECAI}, keywords = {isac-www}, pages = {1–6}, title = {Active Bird2Vec: Towards End-To-End Bird Sound Monitoring with Transformers}, year = 2023 }
Self-awareness in Cyber-Physical Systems: Recent Developments and Open Challenges. Esterle, Lukas; Dutt, Nikil; Gruhl, Christian; Lewis, Peter R.; Marcenaro, Lucio; Regazzoni, Carlo; Jantsch, Axel. In Design, Automation & Test in Europe Conference & Exhibition (DATE), bll 1–6. IEEE, 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Self-aware computing systems enable computing systems to reflect on their actions and behavior. This becomes even more relevant in Cyber-Physical Systems where computing systems have to control and interact with elements in the real world. This paper reports on recent advances made in computational self-awareness for cyber-physical systems.
@inproceedings{esterle2023self, abstract = {Self-aware computing systems enable computing systems to reflect on their actions and behavior. This becomes even more relevant in Cyber-Physical Systems where computing systems have to control and interact with elements in the real world. This paper reports on recent advances made in computational self-awareness for cyber-physical systems.}, author = {Esterle, Lukas and Dutt, Nikil and Gruhl, Christian and Lewis, Peter R. and Marcenaro, Lucio and Regazzoni, Carlo and Jantsch, Axel}, booktitle = {Design, Automation & Test in Europe Conference & Exhibition (DATE)}, keywords = {isac-www}, pages = {1–6}, publisher = {IEEE}, title = {Self-awareness in Cyber-Physical Systems: Recent Developments and Open Challenges}, year = 2023 }
Self-Integration and Agent Compatibility. Gruhl, Christian; Sick, Bernhard. In Autonomic Computing and Self-Organizing Systems Companion (ACSOS-C), bll 71–73. IEEE, 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This position paper addresses the issue of agent compatibility in the context of self-aware microsystems (SAMs), which are SISSY systems with self-awareness capabilities. SAMs are designed to adapt their behaviour and collaborate with other agents in multi-agent settings. Achieving compatibility and effective collaboration between agents pose significant challenges, such as detecting incompatibility and discerning it from malicious behaviour. We present our concept of agent compatibility, highlight future challenges, and propose directions for future research in this area. By focusing on agent compatibility, we aim to advance the understanding of SAMs and promote successful collaboration among agents.
@inproceedings{gruhl2023self, abstract = {This position paper addresses the issue of agent compatibility in the context of self-aware microsystems (SAMs), which are SISSY systems with self-awareness capabilities. SAMs are designed to adapt their behaviour and collaborate with other agents in multi-agent settings. Achieving compatibility and effective collaboration between agents pose significant challenges, such as detecting incompatibility and discerning it from malicious behaviour. We present our concept of agent compatibility, highlight future challenges, and propose directions for future research in this area. By focusing on agent compatibility, we aim to advance the understanding of SAMs and promote successful collaboration among agents.}, author = {Gruhl, Christian and Sick, Bernhard}, booktitle = {Autonomic Computing and Self-Organizing Systems Companion (ACSOS-C)}, keywords = {isac-www}, pages = {71–73}, publisher = {IEEE}, title = {Self-Integration and Agent Compatibility}, year = 2023 }
Resilience of Time-Varying Communication Graphs for Consensus of Changing Sets of Computing Agents. Schmidtke, V.; Liu, Z.; Stursberg, O. In 62nd IEEE Conference on Decision and Control, bll 3474–3479. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
System performance of distributed control systems and networked computing systems is strongly dependent of the underlying communication topology. This paper considers the rarely studied problem of how the topology can maintain resilience by reconfiguration in case that agents leave or join the network during online operation. Existing optimization-based approaches which reconfigure the entire network can typically not be used in this case, since the computational burden for online application is too high. Thus, this paper proposes a novel combined offline-online scheme which optimizes the topology for high convergence rate (of e.g. consensus problems) while providing guarantees for the robustness against agent failures. In the offline part, an optimization of the entire topology is carried out using novel constraints to prepare resilience of the online procedure. For the latter, the proposed scheme guarantees that robustness is maintained for joining agents and if a specified number of agents leave the network. In simulation, the proposed scheme is compared to existing approaches and the advantages of the online-offline procedure are demonstrated.
@article{schmidtke2023resilience, abstract = {System performance of distributed control systems and networked computing systems is strongly dependent of the underlying communication topology. This paper considers the rarely studied problem of how the topology can maintain resilience by reconfiguration in case that agents leave or join the network during online operation. Existing optimization-based approaches which reconfigure the entire network can typically not be used in this case, since the computational burden for online application is too high. Thus, this paper proposes a novel combined offline-online scheme which optimizes the topology for high convergence rate (of e.g. consensus problems) while providing guarantees for the robustness against agent failures. In the offline part, an optimization of the entire topology is carried out using novel constraints to prepare resilience of the online procedure. For the latter, the proposed scheme guarantees that robustness is maintained for joining agents and if a specified number of agents leave the network. In simulation, the proposed scheme is compared to existing approaches and the advantages of the online-offline procedure are demonstrated.}, author = {Schmidtke, V. and Liu, Z. and Stursberg, O.}, journal = {62nd IEEE Conference on Decision and Control}, keywords = {isac-www}, pages = {3474-3479}, title = {Resilience of Time-Varying Communication Graphs for Consensus of Changing Sets of Computing Agents}, year = 2023 }
Active Learning with Fast Model Updates and Class-Balanced Selection for Imbalanced Datasets. Huang, Zhixin; He, Yujiang; Herde, Marek; Huseljic, Denis; Sick, Bernhard. In Workshop on Interactive Adapative Learning (IAL), ECML PKDD, bll 28–45. 2023.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Active learning strategies aim to minimize the number of queried samples for model training. However, two challenges in pool-based deep active learning on imbalanced datasets are observed in experiments: (1) the declining performance of active learning strategies due to imbalanced class distribution; (2)~the lack of sample diversity in acquisition batches due to the absence of timely model updates. This paper proposes the AL-FaMoUS, a general solution combining fast model updates and class-balanced minibatch selection to the active learning process. Furthermore, a simplification of the AL-FaMoUS, which selects one single sample in each acquisition minibatch, is experimentally evaluated on four image and three time-series imbalanced datasets. The results demonstrate that the implemented AL-FaMoUS outperforms the other adopted AL strategies, including uncertainty sampling and BALD solely combined with either the fast model update or the class balance selection strategy, in terms of accuracy and Macro F1 score.
@inproceedings{huang2023active, abstract = {Active learning strategies aim to minimize the number of queried samples for model training. However, two challenges in pool-based deep active learning on imbalanced datasets are observed in experiments: (1) the declining performance of active learning strategies due to imbalanced class distribution; (2) the lack of sample diversity in acquisition batches due to the absence of timely model updates. This paper proposes the AL-FaMoUS, a general solution combining fast model updates and class-balanced minibatch selection to the active learning process. Furthermore, a simplification of the AL-FaMoUS, which selects one single sample in each acquisition minibatch, is experimentally evaluated on four image and three time-series imbalanced datasets. The results demonstrate that the implemented AL-FaMoUS outperforms the other adopted AL strategies, including uncertainty sampling and BALD solely combined with either the fast model update or the class balance selection strategy, in terms of accuracy and Macro F1 score.}, author = {Huang, Zhixin and He, Yujiang and Herde, Marek and Huseljic, Denis and Sick, Bernhard}, booktitle = {Workshop on Interactive Adapative Learning (IAL), ECML PKDD}, keywords = {ClassBalancedSelection}, pages = {28–45}, title = {Active Learning with Fast Model Updates and Class-Balanced Selection for Imbalanced Datasets}, year = 2023 }
Context-aware recommendations for extended electric vehicle battery lifetime. Eider, Markus; Sick, Bernhard; Berl, Andreas. In Sustainable Computing: Informatics and Systems (SUSCOM), 37, bl 100845. Elsevier, 2023.
- [ Abstract ]
- [ BibTeX ]
Electric vehicles are a means of reducing CO2 emissions in transportation. However, the sustainability of electric vehicle batteries is affected by battery health degradation, which decreases their overall lifetime. This results in a substantial amount of depleted batteries due to replacements. Users have a major impact on battery health degradation through their actions while handling electric vehicles, such as the use of fast charging. To mitigate this problem, this article presents a methodology to generate user guidance for batteryfriendly actions in the upcoming use. Therefore, we first identify general recommendations from related work and combine them with the vehicle context in order to define context-aware recommendations in the form of if–then rules. These context-aware recommendations are then used to generate user advice. Second, the article covers how to predict the vehicle context in order to determine necessary recommendations. Third, a prescriptive recommendation system architecture is proposed, which takes vehicle context information, and produces user guidance. Finally, we test the proposed architecture using fuzzy logic as decision system. Overall, the architecture provides satisfactory user advice
@article{eider2023context, abstract = {Electric vehicles are a means of reducing CO2 emissions in transportation. However, the sustainability of electric vehicle batteries is affected by battery health degradation, which decreases their overall lifetime. This results in a substantial amount of depleted batteries due to replacements. Users have a major impact on battery health degradation through their actions while handling electric vehicles, such as the use of fast charging. To mitigate this problem, this article presents a methodology to generate user guidance for batteryfriendly actions in the upcoming use. Therefore, we first identify general recommendations from related work and combine them with the vehicle context in order to define context-aware recommendations in the form of if–then rules. These context-aware recommendations are then used to generate user advice. Second, the article covers how to predict the vehicle context in order to determine necessary recommendations. Third, a prescriptive recommendation system architecture is proposed, which takes vehicle context information, and produces user guidance. Finally, we test the proposed architecture using fuzzy logic as decision system. Overall, the architecture provides satisfactory user advice}, author = {Eider, Markus and Sick, Bernhard and Berl, Andreas}, journal = {Sustainable Computing: Informatics and Systems (SUSCOM)}, keywords = {isac-www}, pages = 100845, publisher = {Elsevier}, title = {Context-aware recommendations for extended electric vehicle battery lifetime}, volume = 37, year = 2023 }

2022

Stream-based active learning for sliding windows under the influence of verification latency. Pham, Tuan; Kottke, Daniel; Krempl, Georg; Sick, Bernhard. In Machine Learning, 111(6), bll 2011–2036. Springer, 2022.
- [ Abstract ]
- [ BibTeX ]
Stream-based active learning (AL) strategies minimize the labeling effort by querying labels that improve the classifier's performance the most. So far, these strategies neglect the fact that an oracle or expert requires time to provide a queried label. We show that existing AL methods deteriorate or even fail under the influence of such verification latency. The problem with these methods is that they estimate a label's utility on the currently available labeled data. However, when this label would arrive, some of the current data may have gotten outdated and new labels have arrived. In this article, we propose to simulate the available data at the time when the label would arrive. Therefore, our method Forgetting and Simulating (FS) forgets outdated information and simulates the delayed labels to get more realistic utility estimates. We assume to know the label's arrival date a priori and the classifier's training data to be bounded by a sliding window. Our extensive experiments show that FS improves stream-based AL strategies in settings with both, constant and variable verification latency.
@article{pham2022stream, abstract = {Stream-based active learning (AL) strategies minimize the labeling effort by querying labels that improve the classifier's performance the most. So far, these strategies neglect the fact that an oracle or expert requires time to provide a queried label. We show that existing AL methods deteriorate or even fail under the influence of such verification latency. The problem with these methods is that they estimate a label's utility on the currently available labeled data. However, when this label would arrive, some of the current data may have gotten outdated and new labels have arrived. In this article, we propose to simulate the available data at the time when the label would arrive. Therefore, our method Forgetting and Simulating (FS) forgets outdated information and simulates the delayed labels to get more realistic utility estimates. We assume to know the label's arrival date a priori and the classifier's training data to be bounded by a sliding window. Our extensive experiments show that FS improves stream-based AL strategies in settings with both, constant and variable verification latency.}, author = {Pham, Tuan and Kottke, Daniel and Krempl, Georg and Sick, Bernhard}, journal = {Machine Learning}, keywords = {isac-www}, number = 6, pages = {2011–2036}, publisher = {Springer}, title = {Stream-based active learning for sliding windows under the influence of verification latency}, volume = 111, year = 2022 }
Active Learning in Multivariate Time Series Anomaly Detection. Huang, Zhixin. In Organic Computing -- Doctoral Dissertation Colloquium 2021, S. Tomforde, C. Krupitzer (reds.), bll 113–124. kassel university press, 2022.
- [ Abstract ]
- [ BibTeX ]
Multivariate time series anomaly detection is an important research field of industry, science, and finances. In practical applications, unlabeled data is abundant, and labeled data is rare. Experts’ experience plays a key role in data labeling, while experts often have limited time, are expensive, and are concerned primarily with finding the potential anomalies. Active learning, which has been widely used to select the most informative data for labeling through experts, can improve training with less labeling effort. However, most active learning algorithms have not been evaluated on datasets with imbalanced class frequencies, such as anomaly detection. Although some semi-supervised and active learning methods were proposed to handle univariate time series data, not much work has been done on multivariate time series classification. Therefore, this article identifies critical challenges and proposes potential solutions for active learning in multivariate time series anomaly detection. We combine all possible solutions into a framework that conforms to the scope of the observation/control system in organic computing.
@incollection{huang2022active, abstract = {Multivariate time series anomaly detection is an important research field of industry, science, and finances. In practical applications, unlabeled data is abundant, and labeled data is rare. Experts’ experience plays a key role in data labeling, while experts often have limited time, are expensive, and are concerned primarily with finding the potential anomalies. Active learning, which has been widely used to select the most informative data for labeling through experts, can improve training with less labeling effort. However, most active learning algorithms have not been evaluated on datasets with imbalanced class frequencies, such as anomaly detection. Although some semi-supervised and active learning methods were proposed to handle univariate time series data, not much work has been done on multivariate time series classification. Therefore, this article identifies critical challenges and proposes potential solutions for active learning in multivariate time series anomaly detection. We combine all possible solutions into a framework that conforms to the scope of the observation/control system in organic computing.}, author = {Huang, Zhixin}, booktitle = {Organic Computing – Doctoral Dissertation Colloquium 2021}, editor = {Tomforde, Sven and Krupitzer, Christian}, keywords = {isac-www}, pages = {113–124}, publisher = {kassel university press}, title = {Active Learning in Multivariate Time Series Anomaly Detection}, year = 2022 }
Efficient SVDD sampling with approximation guarantees for the decision boundary. Englhardt, Adrian; Trittenbach, Holger; Kottke, Daniel; Sick, Bernhard; Böhm, Klemens. In Machine Learning, 111(4), bll 1349–1375. Springer, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Support Vector Data Description (SVDD) is a popular one-class classifier for anomaly and novelty detection. But despite its effectiveness, SVDD does not scale well with data size. To avoid prohibitive training times, sampling methods select small subsets of the training data on which SVDD trains a decision boundary hopefully equivalent to the one obtained on the full data set. According to the literature, a good sample should therefore contain so-called boundary observations that SVDD would select as support vectors on the full data set. However, non-boundary observations also are essential to not fragment contiguous inlier regions and avoid poor classification accuracy. Other aspects, such as selecting a sufficiently representative sample, are important as well. But existing sampling methods largely overlook them, resulting in poor classification accuracy. In this article, we study how to select a sample considering these points. Our approach is to frame SVDD sampling as an optimization problem, where constraints guarantee that sampling indeed approximates the original decision boundary. We then propose RAPID, an efficient algorithm to solve this optimization problem. RAPID does not require any tuning of parameters, is easy to implement and scales well to large data sets. We evaluate our approach on real-world and synthetic data. Our evaluation is the most comprehensive one for SVDD sampling so far. Our results show that RAPID outperforms its competitors in classification accuracy, in sample size, and in runtime.
@article{englhardt2022efficient, abstract = {Support Vector Data Description (SVDD) is a popular one-class classifier for anomaly and novelty detection. But despite its effectiveness, SVDD does not scale well with data size. To avoid prohibitive training times, sampling methods select small subsets of the training data on which SVDD trains a decision boundary hopefully equivalent to the one obtained on the full data set. According to the literature, a good sample should therefore contain so-called boundary observations that SVDD would select as support vectors on the full data set. However, non-boundary observations also are essential to not fragment contiguous inlier regions and avoid poor classification accuracy. Other aspects, such as selecting a sufficiently representative sample, are important as well. But existing sampling methods largely overlook them, resulting in poor classification accuracy. In this article, we study how to select a sample considering these points. Our approach is to frame SVDD sampling as an optimization problem, where constraints guarantee that sampling indeed approximates the original decision boundary. We then propose RAPID, an efficient algorithm to solve this optimization problem. RAPID does not require any tuning of parameters, is easy to implement and scales well to large data sets. We evaluate our approach on real-world and synthetic data. Our evaluation is the most comprehensive one for SVDD sampling so far. Our results show that RAPID outperforms its competitors in classification accuracy, in sample size, and in runtime.}, author = {Englhardt, Adrian and Trittenbach, Holger and Kottke, Daniel and Sick, Bernhard and Böhm, Klemens}, journal = {Machine Learning}, keywords = {isac-www}, number = 4, pages = {1349–1375}, publisher = {Springer}, title = {Efficient {SVDD} sampling with approximation guarantees for the decision boundary}, volume = 111, year = 2022 }
Graph Neural Networks Designed for Different Graph Types: A Survey. Thomas, Josephine M.; Moallemy-Oureh, Alice; Beddar-Wiesing, Silvia; Holzhüter, Clara. In arXiv e-prints, bl arXiv:2204.03080. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Graphs are ubiquitous in nature and can therefore serve as models for many practical but also theoretical problems. Based on this, the young research field of Graph Neural Networks (GNNs) has emerged. Despite the youth of the field and the speed in which new models are developed, many good surveys have been published in the last years. Nevertheless, an overview on which graph types can be modeled by GNNs is missing. In this survey, we give a detailed overview of already existing GNNs and, unlike previous surveys, categorize them according to their ability to handle different graph types and properties. We consider GNNs operating on static as well as on dynamic graphs of different structural constitutions, with or without node or edge attributes. Moreover in the dynamic case, we separate the models in discrete-time and continuous-time dynamic graphs based on their architecture. While ordering the existing GNN models, we find, that there are still graph types, that are not or only rarely covered by existing GNN models. We point out where models are missing and give potential reasons for their absence.
@article{thomas2022graph, abstract = {Graphs are ubiquitous in nature and can therefore serve as models for many practical but also theoretical problems. Based on this, the young research field of Graph Neural Networks (GNNs) has emerged. Despite the youth of the field and the speed in which new models are developed, many good surveys have been published in the last years. Nevertheless, an overview on which graph types can be modeled by GNNs is missing. In this survey, we give a detailed overview of already existing GNNs and, unlike previous surveys, categorize them according to their ability to handle different graph types and properties. We consider GNNs operating on static as well as on dynamic graphs of different structural constitutions, with or without node or edge attributes. Moreover in the dynamic case, we separate the models in discrete-time and continuous-time dynamic graphs based on their architecture. While ordering the existing GNN models, we find, that there are still graph types, that are not or only rarely covered by existing GNN models. We point out where models are missing and give potential reasons for their absence.}, author = {Thomas, Josephine M. and Moallemy-Oureh, Alice and Beddar-Wiesing, Silvia and Holzhüter, Clara}, journal = {arXiv e-prints}, keywords = {Survey}, pages = {arXiv:2204.03080}, title = {Graph Neural Networks Designed for Different Graph Types: A Survey}, year = 2022 }
Predicting flow stress behavior of an AA7075 alloy using machine learning methods. Decke, Jens; Engelhardt, Anna; Rauch, Lukas; Degener, Sebastian; Sajjadifar, Seyedvahid; Scharifi, Emad; Steinhoff, Kurt; Niendorf, Thomas; Sick, Bernhard. In Crystals, 9(12), bll 1–19. MDPI, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This work focuses on the prediction of hot deformation behavior of thermo-mechanically processed precipitation hardenable aluminum alloy AA7075. Data available are focusing on a novel hot forming process at different tool temperatures ranging from 24°C to 350°C to set different cooling rates after solution-heat-treatment. Isothermal uniaxial tensile tests in the temperature range from 200°C to 400°C and at strain rates ranging from 0.001 s^-1 to 0.1 s^-1 were carried out on four different material conditions. The present paper mainly focuses on a comparative study of modeling techniques based on Machine Learning (ML) and the well-known Zerilli-Armstrong model (Z-A) as an empirically based reference. Work focused on predicting single data points of curves that the model was trained on. Due to the novel way data were split with respect to training and testing data, it becomes possible to predict entire stress-strain curves which leads to a reduction in the number of required laboratory experiments, finally saving costs and time in future experiments. While all investigated ML methods showed a higher performance than the Z-A model, the extreme Gradient Boosting model (XGB) showed the superior results, i.e., highest error reduction of 91% with respect to the Mean Squared Error.
@article{decke2022predicting, abstract = {This work focuses on the prediction of hot deformation behavior of thermo-mechanically processed precipitation hardenable aluminum alloy AA7075. Data available are focusing on a novel hot forming process at different tool temperatures ranging from 24°C to 350°C to set different cooling rates after solution-heat-treatment. Isothermal uniaxial tensile tests in the temperature range from 200°C to 400°C and at strain rates ranging from 0.001 s^{-1} to 0.1 s^{-1} were carried out on four different material conditions. The present paper mainly focuses on a comparative study of modeling techniques based on Machine Learning (ML) and the well-known Zerilli-Armstrong model (Z-A) as an empirically based reference. Work focused on predicting single data points of curves that the model was trained on. Due to the novel way data were split with respect to training and testing data, it becomes possible to predict entire stress-strain curves which leads to a reduction in the number of required laboratory experiments, finally saving costs and time in future experiments. While all investigated ML methods showed a higher performance than the Z-A model, the extreme Gradient Boosting model (XGB) showed the superior results, i.e., highest error reduction of 91% with respect to the Mean Squared Error.}, author = {Decke, Jens and Engelhardt, Anna and Rauch, Lukas and Degener, Sebastian and Sajjadifar, Seyedvahid and Scharifi, Emad and Steinhoff, Kurt and Niendorf, Thomas and Sick, Bernhard}, journal = {Crystals}, keywords = {AA7075}, number = 12, pages = {1–19}, publisher = {MDPI}, title = {Predicting flow stress behavior of an AA7075 alloy using machine learning methods}, volume = 9, year = 2022 }
Fast Bayesian Updates for Deep Learning with a Use Case in Active Learning. Herde, Marek; Huang, Zhixin; Huseljic, Denis; Kottke, Daniel; Vogt, Stephan; Sick, Bernhard. In arXiv e-prints, bl arXiv:2210.06112. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Retraining deep neural networks when new data arrives is typically computationally expensive. Moreover, certain applications do not allow such costly retraining due to time or computational constraints. Fast Bayesian updates are a possible solution to this issue. Therefore, we propose a Bayesian update based on Monte-Carlo samples and a last-layer Laplace approximation for different Bayesian neural network types, i.e., Dropout, Ensemble, and Spectral Normalized Neural Gaussian Process (SNGP). In a large-scale evaluation study, we show that our updates combined with SNGP represent a fast and competitive alternative to costly retraining. As a use case, we combine the Bayesian updates for SNGP with different sequential query strategies to exemplarily demonstrate their improved selection performance in active learning.
@article{herde2022fast, abstract = {Retraining deep neural networks when new data arrives is typically computationally expensive. Moreover, certain applications do not allow such costly retraining due to time or computational constraints. Fast Bayesian updates are a possible solution to this issue. Therefore, we propose a Bayesian update based on Monte-Carlo samples and a last-layer Laplace approximation for different Bayesian neural network types, i.e., Dropout, Ensemble, and Spectral Normalized Neural Gaussian Process (SNGP). In a large-scale evaluation study, we show that our updates combined with SNGP represent a fast and competitive alternative to costly retraining. As a use case, we combine the Bayesian updates for SNGP with different sequential query strategies to exemplarily demonstrate their improved selection performance in active learning.}, author = {Herde, Marek and Huang, Zhixin and Huseljic, Denis and Kottke, Daniel and Vogt, Stephan and Sick, Bernhard}, journal = {arXiv e-prints}, keywords = {ActiveLearning}, pages = {arXiv:2210.06112}, title = {Fast Bayesian Updates for Deep Learning with a Use Case in Active Learning}, year = 2022 }
FDGNN: Fully Dynamic Graph Neural Network. Moallemy-Oureh, Alice; Beddar-Wiesing, Silcia; Nather, Rüdiger; Thomas, Josephine M. In arXiv e-prints, bl arXiv:2206.03469. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Dynamic Graph Neural Networks recently became more and more important as graphs from many scientific fields, ranging from mathematics, biology, social sciences, and physics to computer science, are dynamic by nature. While temporal changes (dynamics) play an essential role in many real-world applications, most of the models in the literature on Graph Neural Networks (GNN) process static graphs. The few GNN models on dynamic graphs only consider exceptional cases of dynamics, e.g., node attribute-dynamic graphs or structure-dynamic graphs limited to additions or changes to the graph's edges, etc. Therefore, we present a novel Fully Dynamic Graph Neural Network (FDGNN) that can handle fully-dynamic graphs in continuous time. The proposed method provides a node and an edge embedding that includes their activity to address added and deleted nodes or edges, and possible attributes. Furthermore, the embeddings specify Temporal Point Processes for each event to encode the distributions of the structure- and attribute-related incoming graph events. In addition, our model can be updated efficiently by considering single events for local retraining.
@article{moallemyoureh2022fdgnn, abstract = {Dynamic Graph Neural Networks recently became more and more important as graphs from many scientific fields, ranging from mathematics, biology, social sciences, and physics to computer science, are dynamic by nature. While temporal changes (dynamics) play an essential role in many real-world applications, most of the models in the literature on Graph Neural Networks (GNN) process static graphs. The few GNN models on dynamic graphs only consider exceptional cases of dynamics, e.g., node attribute-dynamic graphs or structure-dynamic graphs limited to additions or changes to the graph's edges, etc. Therefore, we present a novel Fully Dynamic Graph Neural Network (FDGNN) that can handle fully-dynamic graphs in continuous time. The proposed method provides a node and an edge embedding that includes their activity to address added and deleted nodes or edges, and possible attributes. Furthermore, the embeddings specify Temporal Point Processes for each event to encode the distributions of the structure- and attribute-related incoming graph events. In addition, our model can be updated efficiently by considering single events for local retraining.}, author = {Moallemy-Oureh, Alice and Beddar-Wiesing, Silcia and Nather, Rüdiger and Thomas, Josephine M.}, journal = {arXiv e-prints}, keywords = {isac-www}, pages = {arXiv:2206.03469}, title = {FDGNN: Fully Dynamic Graph Neural Network}, year = 2022 }
A Review of Uncertainty Calibration in Pretrained Object Detectors. Huseljic, Denis; Herde, Marek; Muejde, Mehmet; Sick, Bernhard. In arXiv e-prints, bl arXiv:2210.02935. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
In the field of deep learning based computer vision, the development of deep object detection has led to unique paradigms (e.g., two-stage or set-based) and architectures (e.g., Faster-RCNN or DETR) which enable outstanding performance on challenging benchmark datasets. Despite this, the trained object detectors typically do not reliably assess uncertainty regarding their own knowledge, and the quality of their probabilistic predictions is usually poor. As these are often used to make subsequent decisions, such inaccurate probabilistic predictions must be avoided. In this work, we investigate the uncertainty calibration properties of different pretrained object detection architectures in a multi-class setting. We propose a framework to ensure a fair, unbiased, and repeatable evaluation and conduct detailed analyses assessing the calibration under distributional changes (e.g., distributional shift and application to out-of-distribution data). Furthermore, by investigating the influence of different detector paradigms, post-processing steps, and suitable choices of metrics, we deliver novel insights into why poor detector calibration emerges. Based on these insights, we are able to improve the calibration of a detector by simply finetuning its last layer.
@article{huseljic2022review, abstract = {In the field of deep learning based computer vision, the development of deep object detection has led to unique paradigms (e.g., two-stage or set-based) and architectures (e.g., Faster-RCNN or DETR) which enable outstanding performance on challenging benchmark datasets. Despite this, the trained object detectors typically do not reliably assess uncertainty regarding their own knowledge, and the quality of their probabilistic predictions is usually poor. As these are often used to make subsequent decisions, such inaccurate probabilistic predictions must be avoided. In this work, we investigate the uncertainty calibration properties of different pretrained object detection architectures in a multi-class setting. We propose a framework to ensure a fair, unbiased, and repeatable evaluation and conduct detailed analyses assessing the calibration under distributional changes (e.g., distributional shift and application to out-of-distribution data). Furthermore, by investigating the influence of different detector paradigms, post-processing steps, and suitable choices of metrics, we deliver novel insights into why poor detector calibration emerges. Based on these insights, we are able to improve the calibration of a detector by simply finetuning its last layer.}, author = {Huseljic, Denis and Herde, Marek and Muejde, Mehmet and Sick, Bernhard}, journal = {arXiv e-prints}, keywords = {isac-www}, pages = {arXiv:2210.02935}, title = {A Review of Uncertainty Calibration in Pretrained Object Detectors}, year = 2022 }
A Holistic View on Probabilistic Trajectory Forecasting -- Case Study: Cyclist Intention Detection. Zernetsch, Stefan; Reichert, Hannes; Kress, Viktor; Doll, Konrad; Sick, Bernhard. In IEEE Intelligent Vehicles Symposium (IV), bll 265–272. IEEE, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This article presents a holistic approach for probabilistic cyclist intention detection. By combining probabilities for the current cyclist motion states with a probabilistic ensemble trajectory forecast, we are able to generate reliable estimates of cyclists’ future positions. The probabilities are used as weights in the probabilistic ensemble trajectory forecast. The ensemble consists of specialized models, which produce individual forecasts in the form of Gaussian distributions under the assumption of a certain motion state of the cyclist (e.g. cyclist is starting). By weighting the specialized models, we create forecasts in the form of Gaussian mixtures that define regions within which the cyclists will reside with a certain probability. To evaluate our method, we rate the reliability, sharpness, and positional accuracy of our forecasted distributions. We compare our method to a widely used unimodal approach which produces forecasts in the form of Gaussian distributions and show that our method is able to produce more reliable and sharper outputs while retaining comparable positional accuracy. By comparing two different methods for basic movement detection, one based on the past cyclist trajectory and one using past image sequences, we demonstrate that the results can be further improved by incorporating video information. Our methods are evaluated using a dataset created at a public intersection. The code and the dataset are publicly available.
@inproceedings{zernetsch2022holistic, abstract = {This article presents a holistic approach for probabilistic cyclist intention detection. By combining probabilities for the current cyclist motion states with a probabilistic ensemble trajectory forecast, we are able to generate reliable estimates of cyclists’ future positions. The probabilities are used as weights in the probabilistic ensemble trajectory forecast. The ensemble consists of specialized models, which produce individual forecasts in the form of Gaussian distributions under the assumption of a certain motion state of the cyclist (e.g. cyclist is starting). By weighting the specialized models, we create forecasts in the form of Gaussian mixtures that define regions within which the cyclists will reside with a certain probability. To evaluate our method, we rate the reliability, sharpness, and positional accuracy of our forecasted distributions. We compare our method to a widely used unimodal approach which produces forecasts in the form of Gaussian distributions and show that our method is able to produce more reliable and sharper outputs while retaining comparable positional accuracy. By comparing two different methods for basic movement detection, one based on the past cyclist trajectory and one using past image sequences, we demonstrate that the results can be further improved by incorporating video information. Our methods are evaluated using a dataset created at a public intersection. The code and the dataset are publicly available.}, author = {Zernetsch, Stefan and Reichert, Hannes and Kress, Viktor and Doll, Konrad and Sick, Bernhard}, booktitle = {IEEE Intelligent Vehicles Symposium (IV)}, keywords = {isac-www}, pages = {265–272}, publisher = {IEEE}, title = {A Holistic View on Probabilistic Trajectory Forecasting – Case Study: Cyclist Intention Detection}, year = 2022 }
Design of Explainability Module with Experts in the Loop for Visualization and Dynamic Adjustment of Continual Learning. He, Yujiang; Huang, Zhixin; Sick, Bernhard. In Workshop on Interactive Machine Learning Workshop (IMLW), AAAI, bll 1–6. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Continual learning can enable neural networks to evolve by learning new tasks sequentially in task-changing scenarios. However, two general challenges should be overcome in further research before we apply this technique to real-world applications. Firstly, newly collected novelties from the data stream in applications could contain anomalies that are meaningless for continual learning. Instead of viewing them as a new task for updating, we have to filter out such anomalies to reduce the disturbance of extremely high-entropy data for the progression of convergence. Secondly, fewer efforts have been put into research regarding the explainability of continual learning, which leads to a lack of transparency and credibility of the updated neural networks. Elaborated explanations about the process and result of continual learning can help experts in judgment and making decisions. Therefore, we propose the conceptual design of an explainability module with experts in the loop based on techniques, such as dimension reduction, visualization, and evaluation strategies. This work aims to overcome the mentioned challenges by sufficiently explaining and visualizing the identified anomalies and the updated neural network. With the help of this module, experts can be more confident in decision-making regarding anomaly filtering, dynamic adjustment of hyperparameters, data backup, etc.
@inproceedings{he2022design, abstract = {Continual learning can enable neural networks to evolve by learning new tasks sequentially in task-changing scenarios. However, two general challenges should be overcome in further research before we apply this technique to real-world applications. Firstly, newly collected novelties from the data stream in applications could contain anomalies that are meaningless for continual learning. Instead of viewing them as a new task for updating, we have to filter out such anomalies to reduce the disturbance of extremely high-entropy data for the progression of convergence. Secondly, fewer efforts have been put into research regarding the explainability of continual learning, which leads to a lack of transparency and credibility of the updated neural networks. Elaborated explanations about the process and result of continual learning can help experts in judgment and making decisions. Therefore, we propose the conceptual design of an explainability module with experts in the loop based on techniques, such as dimension reduction, visualization, and evaluation strategies. This work aims to overcome the mentioned challenges by sufficiently explaining and visualizing the identified anomalies and the updated neural network. With the help of this module, experts can be more confident in decision-making regarding anomaly filtering, dynamic adjustment of hyperparameters, data backup, etc.}, author = {He, Yujiang and Huang, Zhixin and Sick, Bernhard}, booktitle = {Workshop on Interactive Machine Learning Workshop (IMLW), AAAI}, keywords = {Evaluation}, pages = {1–6}, title = {Design of Explainability Module with Experts in the Loop for Visualization and Dynamic Adjustment of Continual Learning}, year = 2022 }
On Stability of Network Control Systems with Switching Transmission Probabilities. Hahn, Jannik; Stursberg, Olaf. In IFAC-PapersOnLine, 55(13), bll 276–281. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
The paper addresses modeling, stability analysis, and control of a class of networked control systems, in which the communication network, and thus the probability for successful transmission of sensor data, varies over time. This variation is modeled probabilistically by Markov chains with time-varying transition probabilities. The offline synthesis of output feedback controllers for this system class is addressed with focus on criteria for mean square stability.
@article{HAHN2022276, abstract = {The paper addresses modeling, stability analysis, and control of a class of networked control systems, in which the communication network, and thus the probability for successful transmission of sensor data, varies over time. This variation is modeled probabilistically by Markov chains with time-varying transition probabilities. The offline synthesis of output feedback controllers for this system class is addressed with focus on criteria for mean square stability.}, author = {Hahn, Jannik and Stursberg, Olaf}, journal = {IFAC-PapersOnLine}, keywords = {isac-www}, note = {9th IFAC Conference on Networked Systems NECSYS 2022}, number = 13, pages = {276-281}, title = {On Stability of Network Control Systems with Switching Transmission Probabilities}, volume = 55, year = 2022 }
Distributed Solution of Mixed-Integer Programs by ADMM with Closed Duality Gap. Liu, Zonglin; Stursberg, Olaf. In 61st IEEE Conference on Decision and Control (CDC). 2022.
- [ BibTeX ]
@article{liu2022mixedinteger, author = {Liu, Zonglin and Stursberg, Olaf}, journal = {61st IEEE Conference on Decision and Control (CDC)}, keywords = {isac-www}, title = {Distributed Solution of Mixed-Integer Programs by ADMM with Closed Duality Gap}, year = 2022 }
Efficient Fault Detection for Discrete-Time PWA Systems,. Liu, Xinyang; Liu, Zonglin; Wang, Zhenhua; Stursberg, Olaf. In IEEE Control Systems Letters, 6, bll 3361–3366. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This letter considers a class of fault detection problems for discrete-time piecewise-affine (PWA) systems. Inspired by the state-of-the-art of set-membership methods to estimate the state of PWA systems, a novel detection method based on zonotopes and over-approximation of sets is proposed to overcome high computational complexity caused by unknown discrete states. The efficiency and high accuracy of detection of the new method is confirmed through both theoretical analysis and numerical examples.
@article{liu2022efficient, abstract = {This letter considers a class of fault detection problems for discrete-time piecewise-affine (PWA) systems. Inspired by the state-of-the-art of set-membership methods to estimate the state of PWA systems, a novel detection method based on zonotopes and over-approximation of sets is proposed to overcome high computational complexity caused by unknown discrete states. The efficiency and high accuracy of detection of the new method is confirmed through both theoretical analysis and numerical examples.}, author = {Liu, Xinyang and Liu, Zonglin and Wang, Zhenhua and Stursberg, Olaf}, journal = {IEEE Control Systems Letters}, keywords = {isac-www}, pages = {3361-3366}, title = {Efficient Fault Detection for Discrete-Time PWA Systems,}, volume = 6, year = 2022 }
Distributed Optimization for Mixed-Integer Consensus in Multi-Agent Networks. Liu, Zonglin; Stursberg, Olaf. In 2022 European Control Conference (ECC), bll 2196–2202. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This paper considers the consensus of mixed-integer linear programming (MILP) problems as occurring in distributed control and machine learning of multi-agent networks. Unlike existing work on consensus problems, in which the agents only have to agree on the continuous part of their decision variables, this paper proposes a new method to enable them to also agree on the integer part. This mixed-integer setting may arise from distributed control problems of hybrid dynamical systems, or distributed machine learning problems using support vector machines. It is shown in this paper that the consensus of mixed-integer variables is guaranteed to be achieved by a tailored series of continuous consensus problems.
@article{liu2022distributed, abstract = {This paper considers the consensus of mixed-integer linear programming (MILP) problems as occurring in distributed control and machine learning of multi-agent networks. Unlike existing work on consensus problems, in which the agents only have to agree on the continuous part of their decision variables, this paper proposes a new method to enable them to also agree on the integer part. This mixed-integer setting may arise from distributed control problems of hybrid dynamical systems, or distributed machine learning problems using support vector machines. It is shown in this paper that the consensus of mixed-integer variables is guaranteed to be achieved by a tailored series of continuous consensus problems.}, author = {Liu, Zonglin and Stursberg, Olaf}, journal = {2022 European Control Conference (ECC)}, keywords = {isac-www}, pages = {2196-2202}, title = {Distributed Optimization for Mixed-Integer Consensus in Multi-Agent Networks}, year = 2022 }
Actively Controlling and Redesigning Experiments using the Application Case of Free-Electron Laser Pulse Characterization. Dingel, Kristina. In Organic Computing -- Doctoral Dissertation Colloquium 2021, S. Tomforde, C. Krupitzer (reds.), bll 86–98. kassel university press, 2022.
- [ Abstract ]
- [ BibTeX ]
Experiments are the elective tool for developing new technologies, testing hypotheses, and designing new materials in many scientific fields. Thereby, it is state-of-the-art to create experiments traditionally, starting with a hypothesis leading to the product or relevant data after several experiment cycles. There are some techniques such as automation, parallelization, or inverse design to accelerate experiments. Particularly interesting, however, is the approach to integrate immediate data analysis into the experiment cycle. By analyzing experiment output data on-site using, e.g., neural networks, one can make inferences regarding the experiment's current settings at run-time and furthermore improve them. In the following research project, the goal is to develop and incorporate a data analysis stage into an experimental setup, which can be actively adapted and designed on-site. To demonstrate the developed method, the experimental design of a self-amplification of spontaneous emission pulse characterization experiment at a free-electron laser will be used to show to what extent the experiment can be adapted and optimized.
@incollection{dingel2022actively, abstract = {Experiments are the elective tool for developing new technologies, testing hypotheses, and designing new materials in many scientific fields. Thereby, it is state-of-the-art to create experiments traditionally, starting with a hypothesis leading to the product or relevant data after several experiment cycles. There are some techniques such as automation, parallelization, or inverse design to accelerate experiments. Particularly interesting, however, is the approach to integrate immediate data analysis into the experiment cycle. By analyzing experiment output data on-site using, e.g., neural networks, one can make inferences regarding the experiment's current settings at run-time and furthermore improve them. In the following research project, the goal is to develop and incorporate a data analysis stage into an experimental setup, which can be actively adapted and designed on-site. To demonstrate the developed method, the experimental design of a self-amplification of spontaneous emission pulse characterization experiment at a free-electron laser will be used to show to what extent the experiment can be adapted and optimized.}, author = {Dingel, Kristina}, booktitle = {Organic Computing – Doctoral Dissertation Colloquium 2021}, editor = {Tomforde, Sven and Krupitzer, Christian}, keywords = {isac-www}, pages = {86–98}, publisher = {kassel university press}, title = {Actively Controlling and Redesigning Experiments using the Application Case of Free-Electron Laser Pulse Characterization}, year = 2022 }
A Concept for Automated Polarized Web Content Annotation based on Multimodal Active Learning. Herde, Marek; Huseljic, Denis; Mitrovic, Jelena; Granitzer, Michael; Sick, Bernhard. In Workshop on Interactive Adaptive Learning (IAL), ECML PKDD, bll 1–6. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Active learning (AL) techniques hardly cope with complex annotations tasks, where, for example, annotations might express relationships across data modalities. As a use case, we consider the task of automatically detecting and reporting multimodal, polarized web content (PWC). Samples of this content type emerge dynamically, covering a broad spectrum of topics. Thus, training machine learning systems for detecting PWC is challenging, particularly if it needs to be done with minimum annotation cost. In this article, we propose the concept of multimodal AL for complex annotations in the context of PWC detection and formulate the resulting challenges as questions for future research.
@inproceedings{herde2022concept, abstract = {Active learning (AL) techniques hardly cope with complex annotations tasks, where, for example, annotations might express relationships across data modalities. As a use case, we consider the task of automatically detecting and reporting multimodal, polarized web content (PWC). Samples of this content type emerge dynamically, covering a broad spectrum of topics. Thus, training machine learning systems for detecting PWC is challenging, particularly if it needs to be done with minimum annotation cost. In this article, we propose the concept of multimodal AL for complex annotations in the context of PWC detection and formulate the resulting challenges as questions for future research.}, author = {Herde, Marek and Huseljic, Denis and Mitrovic, Jelena and Granitzer, Michael and Sick, Bernhard}, booktitle = {Workshop on Interactive Adaptive Learning (IAL), ECML PKDD}, keywords = {HatefulMemes}, pages = {1–6}, title = {A Concept for Automated Polarized Web Content Annotation based on Multimodal Active Learning}, year = 2022 }
Transfer Learning as an Essential Tool for Digital Twins in Renewable Energy Systems. Nivarthi, Chandana Priya. In Organic Computing -- Doctoral Dissertation Colloquium 2021, S. Tomforde, C. Krupitzer (reds.), bll 47–59. kassel university press, 2022.
- [ Abstract ]
- [ BibTeX ]
Transfer learning (TL), the next frontier in machine learning (ML), has gained much popularity in recent years, due to the various challenges faced in ML, like the requirement of vast amounts of training data, expensive and time-consuming labelling processes for data samples, and long training duration for models. TL is useful in tackling these problems, as it focuses on transferring knowledge from previously solved tasks to new tasks. Digital twins and other intelligent systems need to utilise TL to use the previously gained knowledge and solve new tasks in a more self-reliant way, and to incrementally increase their knowledge base. Therefore, in this article, the critical challenges in power forecasting and anomaly detection in the context of renewable energy systems are identified, and a potential TL framework to meet these challenges is proposed. This article also proposes a feature embedding approach to handle the missing sensors data. The proposed TL methods help to make a system more autonomous in the context of organic computing.
@incollection{nivarthi2022transfer, abstract = {Transfer learning (TL), the next frontier in machine learning (ML), has gained much popularity in recent years, due to the various challenges faced in ML, like the requirement of vast amounts of training data, expensive and time-consuming labelling processes for data samples, and long training duration for models. TL is useful in tackling these problems, as it focuses on transferring knowledge from previously solved tasks to new tasks. Digital twins and other intelligent systems need to utilise TL to use the previously gained knowledge and solve new tasks in a more self-reliant way, and to incrementally increase their knowledge base. Therefore, in this article, the critical challenges in power forecasting and anomaly detection in the context of renewable energy systems are identified, and a potential TL framework to meet these challenges is proposed. This article also proposes a feature embedding approach to handle the missing sensors data. The proposed TL methods help to make a system more autonomous in the context of organic computing.}, author = {Nivarthi, Chandana Priya}, booktitle = {Organic Computing – Doctoral Dissertation Colloquium 2021}, editor = {Tomforde, Sven and Krupitzer, Christian}, keywords = {isac-www}, pages = {47–59}, publisher = {kassel university press}, title = {Transfer Learning as an Essential Tool for Digital Twins in Renewable Energy Systems}, year = 2022 }
Adaptive Explainable Continual Learning Framework for Regression Problems with Focus on Power Forecasts. He, Yujiang. In Organic Computing -- Doctoral Dissertation Colloquium 2021, S. Tomforde, C. Krupitzer (reds.), bll 125–140. kassel university press, 2022.
- [ Abstract ]
- [ BibTeX ]
Compared with traditional deep learning techniques, continual learning enables deep neural networks to learn continually and adaptively. Deep neural networks have to learn unseen tasks and overcome forgetting the knowledge obtained from previously learned tasks as the amount of data keeps increasing in applications. This article proposes two continual learning application scenarios, i.e., the target-domain incremental scenario and the data-domain incremental scenario, to describe the potential challenges in this context. Based on our previous work regarding the CLeaR framework, which is short for continual learning for regression tasks, models will be enabled to extend themselves and to learn data successively. Research topics are related, but not limited, to developing continual deep learning algorithms, strategies for non-stationarity detection in data streams, explainable and visualizable artificial intelligence, etc. Moreover, the framework- and algorithm-related hyperparameters should be dynamically updated in applications. Forecasting experiments will be conducted based on power generation and consumption data collected from real-world applications. A series of comprehensive evaluation metrics and visualization tools are applied to access the experimental results. The proposed framework is expected to be generally applied to other constantly changing scenarios.
@incollection{he2022adaptive, abstract = {Compared with traditional deep learning techniques, continual learning enables deep neural networks to learn continually and adaptively. Deep neural networks have to learn unseen tasks and overcome forgetting the knowledge obtained from previously learned tasks as the amount of data keeps increasing in applications. This article proposes two continual learning application scenarios, i.e., the target-domain incremental scenario and the data-domain incremental scenario, to describe the potential challenges in this context. Based on our previous work regarding the CLeaR framework, which is short for continual learning for regression tasks, models will be enabled to extend themselves and to learn data successively. Research topics are related, but not limited, to developing continual deep learning algorithms, strategies for non-stationarity detection in data streams, explainable and visualizable artificial intelligence, etc. Moreover, the framework- and algorithm-related hyperparameters should be dynamically updated in applications. Forecasting experiments will be conducted based on power generation and consumption data collected from real-world applications. A series of comprehensive evaluation metrics and visualization tools are applied to access the experimental results. The proposed framework is expected to be generally applied to other constantly changing scenarios.}, author = {He, Yujiang}, booktitle = {Organic Computing – Doctoral Dissertation Colloquium 2021}, editor = {Tomforde, Sven and Krupitzer, Christian}, keywords = {isac-www}, pages = {125–140}, publisher = {kassel university press}, title = {Adaptive Explainable Continual Learning Framework for Regression Problems with Focus on Power Forecasts}, year = 2022 }
Stream-Based Active Learning in Changing Environments under Verification Latency. Pham, Tuan. In Organic Computing -- Doctoral Dissertation Colloquium 2021, S. Tomforde, C. Krupitzer (reds.), bll 152–164. kassel university press, 2022.
- [ Abstract ]
- [ BibTeX ]
Labeling data often requires human interaction, which is slow and costly depending on the required expertise. Stream-based active learning reduces the cost of labeling for data stream mining by only querying the labels for instances that improve the classifier the most. However, state-of-the-art stream-based active learning strategies assume that verification latency, i.e., the time between a query and the availability of the queried label, is negligible. In reality, this is usually not the case as the experts who provide such labels are not available all the time, and the labeling process itself requires time. This article investigates the impact of verification latency within a stream-based active learning scenario for changing data streams. We focus on two different types of changes that occur over time, i.e., changes in the number of features and changes in the distribution of instances and/or labels. In particular, we identify challenges that occur due to verification latency and propose ideas to overcome these problems within a stream-based active learning strategy.
@incollection{pham2022stream-based, abstract = {Labeling data often requires human interaction, which is slow and costly depending on the required expertise. Stream-based active learning reduces the cost of labeling for data stream mining by only querying the labels for instances that improve the classifier the most. However, state-of-the-art stream-based active learning strategies assume that verification latency, i.e., the time between a query and the availability of the queried label, is negligible. In reality, this is usually not the case as the experts who provide such labels are not available all the time, and the labeling process itself requires time. This article investigates the impact of verification latency within a stream-based active learning scenario for changing data streams. We focus on two different types of changes that occur over time, i.e., changes in the number of features and changes in the distribution of instances and/or labels. In particular, we identify challenges that occur due to verification latency and propose ideas to overcome these problems within a stream-based active learning strategy.}, author = {Pham, Tuan}, booktitle = {Organic Computing – Doctoral Dissertation Colloquium 2021}, editor = {Tomforde, Sven and Krupitzer, Christian}, keywords = {isac-www}, pages = {152–164}, publisher = {kassel university press}, title = {Stream-Based Active Learning in Changing Environments under Verification Latency}, year = 2022 }
Student Research Abstract: Using Local Activity Encoding for Dynamic Graph Pooling in Stuctural-Dynamic Graphs. Beddar-Wiesing, Silvia. In ACM/SIGAPP Symposium on Applied Computing (SAC), bll 604–609. ACM, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Graphs have recently become more popular for modeling real-world as well as theoretical problems, and several Machine Learning algorithms for learning on them have developed. For graphs that change over time, however, research is still at an early stage. This article proposes a framework DynHeatmaP that generates a constant-size graph sequence representation of graphs whose structure changes over time (structural-dynamic, SDG) in linear time. The framework uses a representation of additions and deletions of nodes and edges (events) to perform dynamic graph pooling. Our algorithm translates the dynamic changes of the nodes and edges to a heatmap representation of the current node activity, regarding additions and deletions in the neighborhood, respectively. The created activity heatmap is utilized as probabilities of the nodes to be sampled in the subsequent pooling step. In addition, the neighborhood of selected nodes has been sampled such that the ratio of additions and deletions is represented accordingly. The obtained downsampled graph sequence can then be used representatively as input to a GNN for learning on the original stream. Focussing on the active regions in a graph, the fast pooling process can reduce the training time significantly and addresses the problem of unbalanced datasets in real-world applications modeled as sparse graphs.
@inproceedings{beddarwiesing2022student, abstract = {Graphs have recently become more popular for modeling real-world as well as theoretical problems, and several Machine Learning algorithms for learning on them have developed. For graphs that change over time, however, research is still at an early stage. This article proposes a framework DynHeatmaP that generates a constant-size graph sequence representation of graphs whose structure changes over time (structural-dynamic, SDG) in linear time. The framework uses a representation of additions and deletions of nodes and edges (events) to perform dynamic graph pooling. Our algorithm translates the dynamic changes of the nodes and edges to a heatmap representation of the current node activity, regarding additions and deletions in the neighborhood, respectively. The created activity heatmap is utilized as probabilities of the nodes to be sampled in the subsequent pooling step. In addition, the neighborhood of selected nodes has been sampled such that the ratio of additions and deletions is represented accordingly. The obtained downsampled graph sequence can then be used representatively as input to a GNN for learning on the original stream. Focussing on the active regions in a graph, the fast pooling process can reduce the training time significantly and addresses the problem of unbalanced datasets in real-world applications modeled as sparse graphs.}, author = {Beddar-Wiesing, Silvia}, booktitle = {ACM/SIGAPP Symposium on Applied Computing (SAC)}, keywords = {isac-www}, pages = {604–609}, publisher = {ACM}, title = {Student Research Abstract: Using Local Activity Encoding for Dynamic Graph Pooling in Stuctural-Dynamic Graphs}, year = 2022 }
Student Research Abstract: Continuous-Time Generative Graph Neural Network for Attributed Dynamic Graphs. Moallemy-Oureh, Alice. In ACM/SIGAPP Symposium on Applied Computing (SAC), bll 600–603. ACM, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
The history of neural networks dates back to the early 1940s and has not only evolved rapidly over time but they are now amongst one of the most popular and powerful machine learning techniques. In this area, graph representation learning (GRL) using graph neural networks (GNNs) has emerged during the early 90s and developed into an impactful approach for modeling graph-structured real-world data, such as social or biological networks. There are a variety of successful applications of GRL in computational neuroscience, chemistry, mathematics, and so on. While temporal changes (dynamics) play an essential role in many real-world applications, most of the literature on GNNs and GRL, deals with static graphs. Since the few GNN models on dynamic graphs only consider exceptional cases of dynamics (i.e., attribute-dynamic graphs in discrete-time representation or structure-dynamic graphs in continuous-time representation), we aim to present a novel GNN model that can handle attribute-dynamic graphs in continuous time. This model learns any kind of graph information embedding, performs node/edge attribute forecasts, and allows for attribute-dynamic graph generation in both discrete and continuous-time.
@inproceedings{moallemyoureh2022student, abstract = {The history of neural networks dates back to the early 1940s and has not only evolved rapidly over time but they are now amongst one of the most popular and powerful machine learning techniques. In this area, graph representation learning (GRL) using graph neural networks (GNNs) has emerged during the early 90s and developed into an impactful approach for modeling graph-structured real-world data, such as social or biological networks. There are a variety of successful applications of GRL in computational neuroscience, chemistry, mathematics, and so on. While temporal changes (dynamics) play an essential role in many real-world applications, most of the literature on GNNs and GRL, deals with static graphs. Since the few GNN models on dynamic graphs only consider exceptional cases of dynamics (i.e., attribute-dynamic graphs in discrete-time representation or structure-dynamic graphs in continuous-time representation), we aim to present a novel GNN model that can handle attribute-dynamic graphs in continuous time. This model learns any kind of graph information embedding, performs node/edge attribute forecasts, and allows for attribute-dynamic graph generation in both discrete and continuous-time.}, author = {Moallemy-Oureh, Alice}, booktitle = {ACM/SIGAPP Symposium on Applied Computing (SAC)}, keywords = {isac-www}, pages = {600–603}, publisher = {ACM}, title = {Student Research Abstract: Continuous-Time Generative Graph Neural Network for Attributed Dynamic Graphs}, year = 2022 }
Using AoI Forecasts in Communicating and Robust Distributed Model-Predictive Control. Hahn, Jannik; Schoeffauer, Richard; Wunder, Gerhard; Stursberg, Olaf. In IEEE Transactions on Control of Network Systems, 9(2), bll 742–752. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
In order to enhance the performance of cyber-physical systems, this paper proposes the integrated design of distributed controllers for distributed plants together with the control of the communication network. Conventional design methods use static interfaces between both entities and therefore rely on worst-case estimations of communication delay, often leading to conservative behavior of the overall system. By contrast, the present approach establishes a robust distributed model-predictive control scheme, in which the local subsystem controllers operate under the assumption of a variable communication schedule that is predicted by a network controller. Using appropriate models for the communication network, the network controller applies a predictive network policy for scheduling the communication among the subsystem controllers across the network. Given the resulting time-varying predictions of the age of information, the paper shows under which conditions the subsystem controllers can robustly stabilize the distributed system. To illustrate the approach, the paper also reports on the application to a vehicle platooning scenario.
@article{hahn2022using, abstract = {In order to enhance the performance of cyber-physical systems, this paper proposes the integrated design of distributed controllers for distributed plants together with the control of the communication network. Conventional design methods use static interfaces between both entities and therefore rely on worst-case estimations of communication delay, often leading to conservative behavior of the overall system. By contrast, the present approach establishes a robust distributed model-predictive control scheme, in which the local subsystem controllers operate under the assumption of a variable communication schedule that is predicted by a network controller. Using appropriate models for the communication network, the network controller applies a predictive network policy for scheduling the communication among the subsystem controllers across the network. Given the resulting time-varying predictions of the age of information, the paper shows under which conditions the subsystem controllers can robustly stabilize the distributed system. To illustrate the approach, the paper also reports on the application to a vehicle platooning scenario.}, author = {Hahn, Jannik and Schoeffauer, Richard and Wunder, Gerhard and Stursberg, Olaf}, journal = {IEEE Transactions on Control of Network Systems}, keywords = {isac-www}, number = 2, pages = {742-752}, title = {Using AoI Forecasts in Communicating and Robust Distributed Model-Predictive Control}, volume = 9, year = 2022 }
Uncertain AoI in stochastic optimal control of constrained LTI systems. Hahn, Jannik; Stursberg, Olaf. In at - Automatisierungstechnik, 70(4), bll 343–354. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This paper addresses finite-time horizon optimal control of control structures with shared communication network. To cope with the uncertainties, induced by network imperfections and exogenous disturbances at the same time, an optimization-based control scheme is proposed. It uses a disturbance feedback and the Age of Information (AoI), a receiver-based measure of communication delays, as central aspects. The disturbance feedback is an extension of the control law used for balanced stochastic optimal control. Balanced optimality is understood as a compromise between minimizing expected deviations from the reference and the minimization of the uncertainty of future states. Time-varying state constraints as well as time-invariant input constraints are considered, and the controllers are synthesized by semi-definite programming.
@article{hahn2022uncertain, abstract = {This paper addresses finite-time horizon optimal control of control structures with shared communication network. To cope with the uncertainties, induced by network imperfections and exogenous disturbances at the same time, an optimization-based control scheme is proposed. It uses a disturbance feedback and the Age of Information (AoI), a receiver-based measure of communication delays, as central aspects. The disturbance feedback is an extension of the control law used for balanced stochastic optimal control. Balanced optimality is understood as a compromise between minimizing expected deviations from the reference and the minimization of the uncertainty of future states. Time-varying state constraints as well as time-invariant input constraints are considered, and the controllers are synthesized by semi-definite programming.}, author = {Hahn, Jannik and Stursberg, Olaf}, journal = {at - Automatisierungstechnik}, keywords = {isac-www}, pages = {343–354}, title = {Uncertain AoI in stochastic optimal control of constrained LTI systems}, volume = {70(4)}, year = 2022 }
Zum strukturvorwissensbasierten Testsignalentwurf mittels Vorsteuerung für die Identifikation von Takagi-Sugeno-Modellen. Himmelsbach, Matthias; Kroll, Andreas. In at -- Automatisierungstechnik, 70(2), bll 119–133. 2022.
- [ Abstract ]
- [ BibTeX ]
In diesem Beitrag wird eine Methode zum Entwurf von Testsignalen für die Identifikation lokal-affiner dynamischer Takagi-Sugeno-SISO-Modelle vorgestellt. Diese Methode verwendet ein zuvor identifizierten Prozessmodell und nutzt so erlangtes Wissen über die Struktur mittels modellprädiktiver Regelung aus. Im Rahmen einer akademischen Fallstudie wird die Methode demonstriert.
@article{2022-MH-at-MPC, abstract = {In diesem Beitrag wird eine Methode zum Entwurf von Testsignalen für die Identifikation lokal-affiner dynamischer Takagi-Sugeno-SISO-Modelle vorgestellt. Diese Methode verwendet ein zuvor identifizierten Prozessmodell und nutzt so erlangtes Wissen über die Struktur mittels modellprädiktiver Regelung aus. Im Rahmen einer akademischen Fallstudie wird die Methode demonstriert.}, author = {Himmelsbach, Matthias and Kroll, Andreas}, journal = {at – Automatisierungstechnik}, keywords = {optimaler}, number = 2, pages = {119–133}, title = {Zum strukturvorwissensbasierten Testsignalentwurf mittels Vorsteuerung für die Identifikation von Takagi-Sugeno-Modellen}, volume = 70, year = 2022 }
A Deep Recurrent Neural Network model for affine quasi-LPV System identification. Rehmer, Alexander; Kroll, Andreas. In Preprints of the 20th European Control Conference (ECC). London, UK, 2022.
- [ BibTeX ]
@inproceedings{RehmerECC2022_2, address = {London, UK}, author = {Rehmer, Alexander and Kroll, Andreas}, booktitle = {Preprints of the 20th European Control Conference (ECC)}, keywords = {mrt,}, month = {12.-15. July}, note = {accepted}, title = {A Deep Recurrent Neural Network model for affine quasi-LPV System identification}, year = 2022 }
Method and Experimental Investigation of Surface Heat Dissipation Measurement using 3D Thermography. Schmoll, Robert; Schramm, Sebastian; Breitenstein, Tom; Kroll, Andreas. In Journal of Sensors and Sensor Systems (JSSS) - Sensors and Measurement Science International (SMSI) 2021 Special Issue, November, bll 41–49. 2022.
- [ BibTeX ]
- [ URL ]
@article{SchmollJSSS2021, author = {Schmoll, Robert and Schramm, Sebastian and Breitenstein, Tom and Kroll, Andreas}, journal = {Journal of Sensors and Sensor Systems (JSSS) - Sensors and Measurement Science International (SMSI) 2021 Special Issue}, keywords = {mrt,}, pages = {41–49}, title = {Method and Experimental Investigation of Surface Heat Dissipation Measurement using 3D Thermography}, volume = {November}, year = 2022 }
Compensating the Size-of-Source Effect: Relationship between the MTF and a Data-Driven Convolution Filter Approach. Schramm, Sebastian; Ebert, Jannik; Schmoll, Robert; Kroll, Andreas. In 16th Quantitative InfraRed Thermography Conference (QIRT). Paris, Frankreich, 2022.
- [ BibTeX ]
@inproceedings{SchrammQIRT2022, address = {Paris, Frankreich}, author = {Schramm, Sebastian and Ebert, Jannik and Schmoll, Robert and Kroll, Andreas}, booktitle = {16th Quantitative InfraRed Thermography Conference (QIRT)}, keywords = {mrt,}, month = {4.-8. Juli}, note = {accepted}, title = {Compensating the Size-of-Source Effect: Relationship between the MTF and a Data-Driven Convolution Filter Approach}, year = 2022 }
Calibration and validation of micromagnetic data for non-destructive analysis of near-surface properties after hard turning. Wegener, Thomas; Liehr, Alexander; Bolender, Artjom; Degener, Sebastian; Wittich, Felix; Kroll, Andreas; Niendorf, Thomas. In HTM Journal of Heat Treatment and Materials. 2022.
- [ BibTeX ]
@article{Wegener_HTM2021, author = {Wegener, Thomas and Liehr, Alexander and Bolender, Artjom and Degener, Sebastian and Wittich, Felix and Kroll, Andreas and Niendorf, Thomas}, booktitle = {HTM Journal of Heat Treatment and Materials}, keywords = {mrt,}, note = {accepted}, title = {Calibration and validation of micromagnetic data for non-destructive analysis of near-surface properties after hard turning}, year = 2022 }
Control of Jump Markov Uncertain Linear Systems With General Probability Distributions. Flues, Patrick; Stursberg, Olaf. In Frontiers of Control Eng., Vol 3. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This paper introduces a method to control a class of jump Markov linear systems with uncertain initialization of the continuous state and affected by disturbances. Both types of uncertainties are modeled as stochastic processes with arbitrarily chosen probability distributions, for which however, the expected values and (co-)variances are known. The paper elaborates on the control task of steering the uncertain system into a target set by use of continuous controls, while chance constraints have to be satisfied for all possible state sequences of the Markov chain. The proposed approach uses a stochastic model predictive control approach on moving finite-time horizons with tailored constraints to achieve the control goal with prescribed confidence. Key steps of the procedure are (i) to over-approximate probabilistic reachable sets by use of the Chebyshev inequality, and (ii) to embed a tightened version of the original constraints into the optimization problem, in order to obtain a control strategy satisfying the specifications. Convergence of the probabilistic reachable sets is attained by suitable bounding of the state covariance matrices for arbitrary Markov chain sequences. The paper presents the main steps of the solution approach, discusses its properties, and illustrates the principle for a numeric example.
@article{flues2022control, abstract = {This paper introduces a method to control a class of jump Markov linear systems with uncertain initialization of the continuous state and affected by disturbances. Both types of uncertainties are modeled as stochastic processes with arbitrarily chosen probability distributions, for which however, the expected values and (co-)variances are known. The paper elaborates on the control task of steering the uncertain system into a target set by use of continuous controls, while chance constraints have to be satisfied for all possible state sequences of the Markov chain. The proposed approach uses a stochastic model predictive control approach on moving finite-time horizons with tailored constraints to achieve the control goal with prescribed confidence. Key steps of the procedure are (i) to over-approximate probabilistic reachable sets by use of the Chebyshev inequality, and (ii) to embed a tightened version of the original constraints into the optimization problem, in order to obtain a control strategy satisfying the specifications. Convergence of the probabilistic reachable sets is attained by suitable bounding of the state covariance matrices for arbitrary Markov chain sequences. The paper presents the main steps of the solution approach, discusses its properties, and illustrates the principle for a numeric example.}, author = {Flues, Patrick and Stursberg, Olaf}, journal = {Frontiers of Control Eng.}, keywords = {isac-www}, title = {Control of Jump Markov Uncertain Linear Systems With General Probability Distributions}, volume = {Vol 3.}, year = 2022 }
Optimizing a superconducting radio-frequency gun using deep reinforcement learning. Meier, David; Ramirez, Luis Vera; Völker, Jens; Viefhaus, Jens; Sick, Bernhard; Hartmann, Gregor. In Physical Review Accelerators and Beams, 25(10), bl 104604. American Physical Society, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Superconducting photoelectron injectors are promising for generating highly brilliant pulsed electron beams with high repetition rates and low emittances. Experiments such as ultrafast electron diffraction, experiments at the Terahertz scale, and energy recovery linac applications require such properties. However, optimizing the beam properties is challenging due to the high number of possible machine parameter combinations. This article shows the successful automated optimization of beam properties utilizing an already existing simulation model. To reduce the required computation time, we replace the costly simulation with a faster approximation with a neural network. For optimization, we propose a reinforcement learning approach leveraging the simple computation of the derivative of the approximation. We prove that our approach outperforms standard optimization methods for the required function evaluations given a defined minimum accuracy.
@article{meier2022optimizing, abstract = {Superconducting photoelectron injectors are promising for generating highly brilliant pulsed electron beams with high repetition rates and low emittances. Experiments such as ultrafast electron diffraction, experiments at the Terahertz scale, and energy recovery linac applications require such properties. However, optimizing the beam properties is challenging due to the high number of possible machine parameter combinations. This article shows the successful automated optimization of beam properties utilizing an already existing simulation model. To reduce the required computation time, we replace the costly simulation with a faster approximation with a neural network. For optimization, we propose a reinforcement learning approach leveraging the simple computation of the derivative of the approximation. We prove that our approach outperforms standard optimization methods for the required function evaluations given a defined minimum accuracy.}, author = {Meier, David and Ramirez, Luis Vera and Völker, Jens and Viefhaus, Jens and Sick, Bernhard and Hartmann, Gregor}, journal = {Physical Review Accelerators and Beams}, keywords = {SuperconductingRF}, number = 10, pages = 104604, publisher = {American Physical Society}, title = {Optimizing a superconducting radio-frequency gun using deep reinforcement learning}, volume = 25, year = 2022 }
Enhancing Active Learning with Weak Supervision and Transfer Learning by Leveraging Information and Knowledge Sources. Rauch, Lukas; Huseljic, Denis; Sick, Bernhard. In Workshop on Interactive Adaptive Learning (IAL), ECML PKDD, bll 27–42. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
One of the major limitations of deploying a machine learning model is the availability of labeled training data and the resulting expensive annotation process. Although active learning (AL) methods may reduce the annotation cost by actively selecting the most-useful instances, a costly human annotator usually provides the labels. Therefore, even with AL, we still consider the annotation process to be time-consuming and expensive. Besides human annotators, though, companies often have a vast amount of information and knowledge sources available that can generate low-cost labels (e.g., a black-box model) or improve the learning process (e.g., a pre-trained model). We present a novel approach that enhances AL with weak supervision (WS) and transfer learning (TL) to reduce the annotation cost by leveraging these sources. Specifically, we consider a black-box model like a rule-based system as an error-prone and weakly-supervised annotator that inexpensively provides labels. We estimate its performance with an annotator model to decide whether a human annotation is required. Additionally, we utilize unlabeled internal and external data by transferring knowledge from a pre-trained model to the AL cycle. We sequentially investigate the impact of WS and TL on annotation cost and model performance in an AL cycle through a use case. Our evaluation shows that our approach can reduce annotation cost by 51% while achieving nearly identical model performance compared to a traditional AL approach.
@inproceedings{rauch2022enhancing, abstract = {One of the major limitations of deploying a machine learning model is the availability of labeled training data and the resulting expensive annotation process. Although active learning (AL) methods may reduce the annotation cost by actively selecting the most-useful instances, a costly human annotator usually provides the labels. Therefore, even with AL, we still consider the annotation process to be time-consuming and expensive. Besides human annotators, though, companies often have a vast amount of information and knowledge sources available that can generate low-cost labels (e.g., a black-box model) or improve the learning process (e.g., a pre-trained model). We present a novel approach that enhances AL with weak supervision (WS) and transfer learning (TL) to reduce the annotation cost by leveraging these sources. Specifically, we consider a black-box model like a rule-based system as an error-prone and weakly-supervised annotator that inexpensively provides labels. We estimate its performance with an annotator model to decide whether a human annotation is required. Additionally, we utilize unlabeled internal and external data by transferring knowledge from a pre-trained model to the AL cycle. We sequentially investigate the impact of WS and TL on annotation cost and model performance in an AL cycle through a use case. Our evaluation shows that our approach can reduce annotation cost by 51% while achieving nearly identical model performance compared to a traditional AL approach.}, author = {Rauch, Lukas and Huseljic, Denis and Sick, Bernhard}, booktitle = {Workshop on Interactive Adaptive Learning (IAL), ECML PKDD}, keywords = {isac-www}, pages = {27–42}, title = {Enhancing Active Learning with Weak Supervision and Transfer Learning by Leveraging Information and Knowledge Sources}, year = 2022 }
The Vision of Self-Management in Cognitive Organic Power Distribution Systems. Loeser, Inga; Braun, Martin; Gruhl, Christian; Menke, Jan-Hendrik; Sick, Bernhard; Tomforde, Sven. In Energies, 15(3), bl 881. MDPI, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Due to the ongoing trend towards a decarbonisation of energy use, the power system is expected to become the backbone of all energy sectors and thus the fundamental critical infrastructure. High penetration with distributed energy resources demands the coordination of a large number of prosumers, partly controlled by home energy management systems (HEMS), to be designed in such a way that the power system’s operational limits are not violated. On the grid level, distribution management systems (DMS) seek to keep the power system in the normal operational state. On the prosumer level, distributed HEMS optimise the internal power flows by setpoint specification of batteries, photovoltaic generators, or flexible loads. The vision of the ODiS (Organic Distribution System) initiative is to develop an architecture to operate a distribution grid reliably, with high resiliency, and fully autonomously by developing “organic” HEMS and DMS which possess multiple self-x capabilities, collectively referred to as self-management. Thus, ODiS seeks answers to the following question: How can we create the most appropriate models, techniques, and algorithms to develop novel kinds of self-configuring, self-organising, self-healing, and self-optimising DMS that are integrally coupled with the distributed HEMS? In this concept paper, the vision of ODiS is presented in detail based on a thorough review of the state of the art.
@article{loeser2022vision, abstract = {Due to the ongoing trend towards a decarbonisation of energy use, the power system is expected to become the backbone of all energy sectors and thus the fundamental critical infrastructure. High penetration with distributed energy resources demands the coordination of a large number of prosumers, partly controlled by home energy management systems (HEMS), to be designed in such a way that the power system’s operational limits are not violated. On the grid level, distribution management systems (DMS) seek to keep the power system in the normal operational state. On the prosumer level, distributed HEMS optimise the internal power flows by setpoint specification of batteries, photovoltaic generators, or flexible loads. The vision of the ODiS (Organic Distribution System) initiative is to develop an architecture to operate a distribution grid reliably, with high resiliency, and fully autonomously by developing “organic” HEMS and DMS which possess multiple self-x capabilities, collectively referred to as self-management. Thus, ODiS seeks answers to the following question: How can we create the most appropriate models, techniques, and algorithms to develop novel kinds of self-configuring, self-organising, self-healing, and self-optimising DMS that are integrally coupled with the distributed HEMS? In this concept paper, the vision of ODiS is presented in detail based on a thorough review of the state of the art.}, author = {Loeser, Inga and Braun, Martin and Gruhl, Christian and Menke, Jan-Hendrik and Sick, Bernhard and Tomforde, Sven}, journal = {Energies}, keywords = {isac-www}, number = 3, pages = 881, publisher = {MDPI}, title = {The Vision of Self-Management in Cognitive Organic Power Distribution Systems}, volume = 15, year = 2022 }
The effect of the forget gate on bifurcation boundaries and dynamics in Recurrent Neural Networks and its implications for gradient-based optimization. Rehmer, Alexander; Kroll, Andras. In Preprints of the International Joint Conference on Neural Networks (IJCNN 2022), bll 1–8. Padua, Italy, 2022.
- [ BibTeX ]
@inproceedings{Rehmer_WCCI_2022, address = {Padua, Italy}, author = {Rehmer, Alexander and Kroll, Andras}, booktitle = {Preprints of the International Joint Conference on Neural Networks (IJCNN 2022)}, keywords = {isac-www}, month = {18.-23. July}, pages = {1-8}, title = {The effect of the forget gate on bifurcation boundaries and dynamics in Recurrent Neural Networks and its implications for gradient-based optimization}, year = 2022 }
Organic Computing -- Doctoral Dissertation Colloquium 2021. Tomforde, Sven; Krupitzer, Christian. In Vol. 20Intelligent Embedded Systems. kassel university press, 2022.
- [ BibTeX ]
@book{tomforde2022organic, editor = {Tomforde, Sven and Krupitzer, Christian}, keywords = {isac-www}, publisher = {kassel university press}, series = {Intelligent Embedded Systems}, title = {Organic Computing – Doctoral Dissertation Colloquium 2021}, volume = 20, year = 2022 }
Self-Adaptive Charging Management in Electric Vehicle Infrastructures based on Reinforcement Learning. Hassouna, Mohamed. In Organic Computing -- Doctoral Dissertation Colloquium 2022, S. Tomforde, C. Krupitzer (reds.), bll 1–16. kassel university press, 2022.
- [ Abstract ]
- [ BibTeX ]
Electromobility plays an increasingly important role in the energy transition. Electric vehicle charging poses challenges to the electricity grid due to high peak demand situations. Therefore, it is crucial to leverage the vehicle's flexibility when charging can be delayed. Conventional charging management systems cannot adapt to frequently changing conditions such as fluctuating renewable energy output, electricity prices and user behavior. In this article, we investigate the application of reinforcement learning approaches for self-adaptive charging management. In particular, we identify challenges regarding realistic environments and adaption to varying topologies and connections among charging stations. We describe related approaches and propose ideas and planned experiments to overcome these problems by utilizing generative models and graph neural networks.
@incollection{hassouna2022self, abstract = {Electromobility plays an increasingly important role in the energy transition. Electric vehicle charging poses challenges to the electricity grid due to high peak demand situations. Therefore, it is crucial to leverage the vehicle's flexibility when charging can be delayed. Conventional charging management systems cannot adapt to frequently changing conditions such as fluctuating renewable energy output, electricity prices and user behavior. In this article, we investigate the application of reinforcement learning approaches for self-adaptive charging management. In particular, we identify challenges regarding realistic environments and adaption to varying topologies and connections among charging stations. We describe related approaches and propose ideas and planned experiments to overcome these problems by utilizing generative models and graph neural networks.}, author = {Hassouna, Mohamed}, booktitle = {Organic Computing – Doctoral Dissertation Colloquium 2022}, editor = {Tomforde, Sven and Krupitzer, Christian}, keywords = {GraphNeuralNetwork}, pages = {1–16}, publisher = {kassel university press}, title = {Self-Adaptive Charging Management in Electric Vehicle Infrastructures based on Reinforcement Learning}, year = 2022 }
Deep Adaptive Knowledge-Aware Learning for E-Mobility. Ali, Mohammad Wazed. In Organic Computing - Doctoral Dissertation Colloquium 2022, S. Tomforde, C. Krupitzer (reds.), bll 139–155. kassel university press, 2022.
- [ Abstract ]
- [ BibTeX ]
Electromobility can reduce global energy demand by reducing the dependence on fossil fuels. Many different types of electric vehicles (EVs) have been produced recently and EV usage is also rising. However, there is no fixed charging behavior of these EV users. Moreover, most renewable energy sources are stochastic in nature. Thus, it is essential that EV charging demand is controlled and adapted in advance to prevent catastrophic failures such as blackouts. Deep learning models are standard technology for power forecasting in many scenarios. However, there is a lack of real-world heterogeneous multi-source data in the e-mobility domain to train a deep learning model from scratch. Moreover, deep learning (DL) models lack model explainability. Therefore, EV stakeholders can benefit from a framework that coordinates deep transfer learning, interactive machine learning, and heterogeneous graph representation learning (to ensure self-explainability). This work aims to develop a novel framework to solve e-mobility-related tasks such as forecasting EV loads efficiently in a resource-constrained environment (data, computing power, etc.). It also describe various research challenges and potential approaches to build the proposed framework. This work also present a deep generative model to synthesize EV charging session data.
@incollection{ali2022deep, abstract = {Electromobility can reduce global energy demand by reducing the dependence on fossil fuels. Many different types of electric vehicles (EVs) have been produced recently and EV usage is also rising. However, there is no fixed charging behavior of these EV users. Moreover, most renewable energy sources are stochastic in nature. Thus, it is essential that EV charging demand is controlled and adapted in advance to prevent catastrophic failures such as blackouts. Deep learning models are standard technology for power forecasting in many scenarios. However, there is a lack of real-world heterogeneous multi-source data in the e-mobility domain to train a deep learning model from scratch. Moreover, deep learning (DL) models lack model explainability. Therefore, EV stakeholders can benefit from a framework that coordinates deep transfer learning, interactive machine learning, and heterogeneous graph representation learning (to ensure self-explainability). This work aims to develop a novel framework to solve e-mobility-related tasks such as forecasting EV loads efficiently in a resource-constrained environment (data, computing power, etc.). It also describe various research challenges and potential approaches to build the proposed framework. This work also present a deep generative model to synthesize EV charging session data.}, author = {Ali, Mohammad Wazed}, booktitle = {Organic Computing - Doctoral Dissertation Colloquium 2022}, editor = {Tomforde, Sven and Krupitzer, Christian}, keywords = {isac-www}, pages = {139–155}, publisher = {kassel university press}, title = {Deep Adaptive Knowledge-Aware Learning for E-Mobility}, year = 2022 }
Pose and Semantic Map Based Probabilistic Forecast of Vulnerable Road Users Trajectories. Kress, Viktor; Jeske, Fabian; Zernetsch, Stefan; Doll, Konrad; Sick, Bernhard. In IEEE Transactions on Intelligent Vehicles, 8(3), bll 2592–2603. IEEE, 2022.
- [ Abstract ]
- [ BibTeX ]
In this article, an approach for probabilistic trajectory forecasting of vulnerable road users (VRUs) is presented, taking into consideration past movements and the surrounding environment. Past movements are represented by 3D poses reflecting the posture and movements of individual body parts. The surrounding environment is modeled in the form of semantic maps showing, e.g., the course of streets, sidewalks, and the occurrence of obstacles. Forecasts are generated in grids discretizing the space and in the form of arbitrary discrete probability distributions. The distributions are evaluated for their reliability, sharpness, and positional accuracy. We compare our method with two approaches providing forecasts in the form of continuous probability distributions, and we discuss their respective advantages and disadvantages. We thereby investigate the impact of poses and semantic maps. Using a technique we refer to as spatial label smoothing, our approach is able to achieve reliable forecasts. Overall, the 3D poses have a positive impact on the forecasts. The semantic maps facilitate the adaptation of the probability distributions to the individual situation and prevent forecasts of trajectories leading through obstacles. Our method is evaluated on a dataset recorded in inner-city traffic using a research vehicle. The dataset has been made publicly available.
@article{kress2022pose, abstract = {In this article, an approach for probabilistic trajectory forecasting of vulnerable road users (VRUs) is presented, taking into consideration past movements and the surrounding environment. Past movements are represented by 3D poses reflecting the posture and movements of individual body parts. The surrounding environment is modeled in the form of semantic maps showing, e.g., the course of streets, sidewalks, and the occurrence of obstacles. Forecasts are generated in grids discretizing the space and in the form of arbitrary discrete probability distributions. The distributions are evaluated for their reliability, sharpness, and positional accuracy. We compare our method with two approaches providing forecasts in the form of continuous probability distributions, and we discuss their respective advantages and disadvantages. We thereby investigate the impact of poses and semantic maps. Using a technique we refer to as spatial label smoothing, our approach is able to achieve reliable forecasts. Overall, the 3D poses have a positive impact on the forecasts. The semantic maps facilitate the adaptation of the probability distributions to the individual situation and prevent forecasts of trajectories leading through obstacles. Our method is evaluated on a dataset recorded in inner-city traffic using a research vehicle. The dataset has been made publicly available.}, author = {Kress, Viktor and Jeske, Fabian and Zernetsch, Stefan and Doll, Konrad and Sick, Bernhard}, journal = {IEEE Transactions on Intelligent Vehicles}, keywords = {isac-www}, number = 3, pages = {2592–2603}, publisher = {IEEE}, title = {Pose and Semantic Map Based Probabilistic Forecast of Vulnerable Road Users Trajectories}, volume = 8, year = 2022 }
Weisfeiler-Lehman goes Dynamic: An Analysis of the Expressive Power of Graph Neural Networks for Attributed and Dynamic Graphs. Beddar-Wiesing, Silvia; D’Inverno, Giuseppe Alessio; Graziani, Caterina; Lachi, Veronica; Moallemy-Oureh, Alice; Scarselli, Franco; Thomas, Josephine. In arXiv e-prints, bl arXiv:2210.03990. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Graph Neural Networks (GNNs) are a large class of relational models for graph processing. Recent theoretical studies on the expressive power of GNNs have focused on two issues. On the one hand, it has been proven that GNNs are as powerful as the Weisfeiler-Lehman test (1-WL) in their ability to distinguish graphs. Moreover, it has been shown that the equivalence enforced by 1-WL equals unfolding equivalence. On the other hand, GNNs turned out to be universal approximators on graphs modulo the constraints enforced by 1-WL/unfolding equivalence. However, these results only apply to Static Undirected Homogeneous Graphs with node attributes. In contrast, real-life applications often involve a variety of graph properties, such as, e.g., dynamics or node and edge attributes. In this paper, we conduct a theoretical analysis of the expressive power of GNNs for these two graph types that are particularly of interest. Dynamic graphs are widely used in modern applications, and its theoretical analysis requires new approaches. The attributed type acts as a standard form for all graph types since it has been shown that all graph types can be transformed without loss to Static Undirected Homogeneous Graphs with attributes on nodes and edges (SAUHG). The study considers generic GNN models and proposes appropriate 1-WL tests for those domains. Then, the results on the expressive power of GNNs are extended by proving that GNNs have the same capability as the 1-WL test in distinguishing dynamic and attributed graphs, the 1-WL equivalence equals unfolding equivalence and that GNNs are universal approximators modulo 1-WL/unfolding equivalence. Moreover, the proof of the approximation capability holds for SAUHGs, which include most of those used in practical applications, and it is constructive in nature allowing to deduce hints on the architecture of GNNs that can achieve the desired accuracy.
@article{beddarwiesing2022weisfeiler, abstract = {Graph Neural Networks (GNNs) are a large class of relational models for graph processing. Recent theoretical studies on the expressive power of GNNs have focused on two issues. On the one hand, it has been proven that GNNs are as powerful as the Weisfeiler-Lehman test (1-WL) in their ability to distinguish graphs. Moreover, it has been shown that the equivalence enforced by 1-WL equals unfolding equivalence. On the other hand, GNNs turned out to be universal approximators on graphs modulo the constraints enforced by 1-WL/unfolding equivalence. However, these results only apply to Static Undirected Homogeneous Graphs with node attributes. In contrast, real-life applications often involve a variety of graph properties, such as, e.g., dynamics or node and edge attributes. In this paper, we conduct a theoretical analysis of the expressive power of GNNs for these two graph types that are particularly of interest. Dynamic graphs are widely used in modern applications, and its theoretical analysis requires new approaches. The attributed type acts as a standard form for all graph types since it has been shown that all graph types can be transformed without loss to Static Undirected Homogeneous Graphs with attributes on nodes and edges (SAUHG). The study considers generic GNN models and proposes appropriate 1-WL tests for those domains. Then, the results on the expressive power of GNNs are extended by proving that GNNs have the same capability as the 1-WL test in distinguishing dynamic and attributed graphs, the 1-WL equivalence equals unfolding equivalence and that GNNs are universal approximators modulo 1-WL/unfolding equivalence. Moreover, the proof of the approximation capability holds for SAUHGs, which include most of those used in practical applications, and it is constructive in nature allowing to deduce hints on the architecture of GNNs that can achieve the desired accuracy.}, author = {Beddar-Wiesing, Silvia and D'Inverno, Giuseppe Alessio and Graziani, Caterina and Lachi, Veronica and Moallemy-Oureh, Alice and Scarselli, Franco and Thomas, Josephine}, journal = {arXiv e-prints}, keywords = {isac-www}, pages = {arXiv:2210.03990}, title = {{W}eisfeiler-{L}ehman goes Dynamic: An Analysis of the Expressive Power of Graph Neural Networks for Attributed and Dynamic Graphs}, year = 2022 }
Unified Autoencoder with Task Embeddings for Multi-Task Learning in Renewable Power Forecasting. Nivarthi, Chandana Priya; Vogt, Stephan; Sick, Bernhard. In International Conference on Machine Learning and Applications (ICMLA), bll 1530–1536. IEEE, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Renewable power generation forecasts using machine learning are typically implemented as single-task learning models, where a separate model is trained for each photovoltaic or wind park. In recent years, transfer learning is gaining popularity in these systems, as it can be used to transfer the knowledge gained from source parks to a target park. However, for transferring the knowledge to a target park, there is a need to determine the most similar source park(s) among the existing parks. This similarity determination using historical power measurements is challenging when the target park has limited to no historical data samples. Therefore, we propose a simple multi-task learning architecture that initially learns a common representation of input weather features among the tasks, using a Unified Autoencoder (UAE) and then learns the task specific information utilizing a Task Embedding layer in a Neural Network (TENN). This proposed architecture, UAE-TENN, can be easily extended to new parks with or without historical data. An elaborate performance comparison of single and multi-task learning models is performed on six photovoltaic and wind farm datasets comprising a total of 529 parks. UAE-TENN significantly improves the performance of power forecasting by 10 to 19% for photovoltaic parks and 5 to 22% for wind parks compared to the baseline models. Even in the zero-shot learning scenario, when there is no historical data, we successfully demonstrate that the UAE-TENN improves the forecast accuracy for a new park by 19% for photovoltaic parks.
@inproceedings{nivarthi2022unified, abstract = {Renewable power generation forecasts using machine learning are typically implemented as single-task learning models, where a separate model is trained for each photovoltaic or wind park. In recent years, transfer learning is gaining popularity in these systems, as it can be used to transfer the knowledge gained from source parks to a target park. However, for transferring the knowledge to a target park, there is a need to determine the most similar source park(s) among the existing parks. This similarity determination using historical power measurements is challenging when the target park has limited to no historical data samples. Therefore, we propose a simple multi-task learning architecture that initially learns a common representation of input weather features among the tasks, using a Unified Autoencoder (UAE) and then learns the task specific information utilizing a Task Embedding layer in a Neural Network (TENN). This proposed architecture, UAE-TENN, can be easily extended to new parks with or without historical data. An elaborate performance comparison of single and multi-task learning models is performed on six photovoltaic and wind farm datasets comprising a total of 529 parks. UAE-TENN significantly improves the performance of power forecasting by 10 to 19% for photovoltaic parks and 5 to 22% for wind parks compared to the baseline models. Even in the zero-shot learning scenario, when there is no historical data, we successfully demonstrate that the UAE-TENN improves the forecast accuracy for a new park by 19% for photovoltaic parks.}, author = {Nivarthi, Chandana Priya and Vogt, Stephan and Sick, Bernhard}, booktitle = {International Conference on Machine Learning and Applications (ICMLA)}, keywords = {isac-www}, pages = {1530–1536}, publisher = {IEEE}, title = {Unified Autoencoder with Task Embeddings for Multi-Task Learning in Renewable Power Forecasting}, year = 2022 }
A Stopping Criterion for Transductive Active Learning. Kottke, Daniel; Sandrock, Christoph; Krempl, Georg; Sick, Bernhard. In European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECMLPKDD), bll 468–484. Springer, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
In transductive active learning, the goal is to determine the correct labels for an unlabeled, known dataset. Therefore, we can either ask an oracle to provide the right label at some cost or use the prediction of a classifier which we train on the labels acquired so far. In contrast, the commonly used (inductive) active learning aims to select instances for labeling out of the unlabeled set to create a generalized classifier, which will be deployed on unknown data. This article formally defines the transductive setting and shows that it requires new solutions. Additionally, we formalize the theoretically cost-optimal stopping point for the transductive scenario. Building upon the probabilistic active learning framework, we propose a new transductive selection strategy that includes a stopping criterion and show its superiority.
@inproceedings{kottke2022stopping, abstract = {In transductive active learning, the goal is to determine the correct labels for an unlabeled, known dataset. Therefore, we can either ask an oracle to provide the right label at some cost or use the prediction of a classifier which we train on the labels acquired so far. In contrast, the commonly used (inductive) active learning aims to select instances for labeling out of the unlabeled set to create a generalized classifier, which will be deployed on unknown data. This article formally defines the transductive setting and shows that it requires new solutions. Additionally, we formalize the theoretically cost-optimal stopping point for the transductive scenario. Building upon the probabilistic active learning framework, we propose a new transductive selection strategy that includes a stopping criterion and show its superiority.}, author = {Kottke, Daniel and Sandrock, Christoph and Krempl, Georg and Sick, Bernhard}, booktitle = {European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECMLPKDD)}, keywords = {isac-www}, pages = {468–484}, publisher = {Springer}, title = {A Stopping Criterion for Transductive Active Learning}, year = 2022 }
Detecting Corner Case in the Context of Highly Automated Driving. Heidecker, Florian. In Organic Computing -- Doctoral Dissertation Colloquium 2021, S. Tomforde, C. Krupitzer (reds.), bll 60–73. kassel university press, 2022.
- [ Abstract ]
- [ BibTeX ]
This article introduces a research proposal that aims to deﬁne, detect, and evaluate corner cases in the context of machine learning (ML) systems and highly automated driving. Corner cases are scenarios or data in a dataset that are recorded only very rarely and therefore only available in small quantities or not at all in the dataset. However, corner cases are very important for model training, as they are needed to improve the model performance in these rare scenarios. On the other hand, in safety- and security-critical applications, such as highly automated driving, corner case data is extremely important for testing and verifying the ML system. The research proposal addresses these challenges and focuses on the deﬁnition and categorization of corner cases. Based on this, ML methods for corner case detection are developed, focusing on object detection for automated driving and trajectory data for intention detection of pedestrians and cyclists. Finally, the research proposal addresses corner case metrics and the evaluation of the developed detection models.
@incollection{florianheidecker2022detecting, abstract = {This article introduces a research proposal that aims to deﬁne, detect, and evaluate corner cases in the context of machine learning (ML) systems and highly automated driving. Corner cases are scenarios or data in a dataset that are recorded only very rarely and therefore only available in small quantities or not at all in the dataset. However, corner cases are very important for model training, as they are needed to improve the model performance in these rare scenarios. On the other hand, in safety- and security-critical applications, such as highly automated driving, corner case data is extremely important for testing and verifying the ML system. The research proposal addresses these challenges and focuses on the deﬁnition and categorization of corner cases. Based on this, ML methods for corner case detection are developed, focusing on object detection for automated driving and trajectory data for intention detection of pedestrians and cyclists. Finally, the research proposal addresses corner case metrics and the evaluation of the developed detection models.}, author = {Heidecker, Florian}, booktitle = {Organic Computing – Doctoral Dissertation Colloquium 2021}, editor = {Tomforde, Sven and Krupitzer, Christian}, keywords = {isac-www}, pages = {60–73}, publisher = {kassel university press}, title = {Detecting Corner Case in the Context of Highly Automated Driving}, year = 2022 }
Space, Time, and Interaction: A Taxonomy of Corner Cases in Trajectory Datasets for Automated Driving. Rösch, Kevin; Heidecker, Florian; Truetsch, Julian; Kowol, Kamil; Schicktanz, Clemens; Bieshaar, Maarten; Sick, Bernhard; Stiller, Christoph. In IEEE Symposium on Computational Intelligence in Vehicles and Transportation Systems (IEEE CIVTS), IEEE SSCI, bll 86–93. IEEE, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Trajectory data analysis is an essential component for highly automated driving. Complex models developed with these data predict other road users’ movement and behavior patterns. Based on these predictions — and additional contextual information such as the course of the road, (traffic) rules, and interaction with other road users — the highly automated vehicle (HAV) must be able to reliably and safely perform the task assigned to it, e.g., moving from point A to B. Ideally, the HAV moves safely through its environment, just as we would expect a human driver to do. However, if unusual trajectories occur, so-called trajectory corner cases, a human driver can usually cope well, but an HAV can quickly get into trouble. In the definition of trajectory corner cases, which we provide in this work, we will consider the relevance of unusual trajectories with respect to the task at hand. Based on this, we will also present a taxonomy of different trajectory corner cases. The categorization of corner cases into the taxonomy will be shown with examples and is done by cause and required data sources. To illustrate the complexity between the machine learning (ML) model and the corner case cause, we present a general processing chain underlying the taxonomy.
@inproceedings{rosch2022space, abstract = {Trajectory data analysis is an essential component for highly automated driving. Complex models developed with these data predict other road users’ movement and behavior patterns. Based on these predictions — and additional contextual information such as the course of the road, (traffic) rules, and interaction with other road users — the highly automated vehicle (HAV) must be able to reliably and safely perform the task assigned to it, e.g., moving from point A to B. Ideally, the HAV moves safely through its environment, just as we would expect a human driver to do. However, if unusual trajectories occur, so-called trajectory corner cases, a human driver can usually cope well, but an HAV can quickly get into trouble. In the definition of trajectory corner cases, which we provide in this work, we will consider the relevance of unusual trajectories with respect to the task at hand. Based on this, we will also present a taxonomy of different trajectory corner cases. The categorization of corner cases into the taxonomy will be shown with examples and is done by cause and required data sources. To illustrate the complexity between the machine learning (ML) model and the corner case cause, we present a general processing chain underlying the taxonomy.}, author = {Rösch, Kevin and Heidecker, Florian and Truetsch, Julian and Kowol, Kamil and Schicktanz, Clemens and Bieshaar, Maarten and Sick, Bernhard and Stiller, Christoph}, booktitle = {IEEE Symposium on Computational Intelligence in Vehicles and Transportation Systems (IEEE CIVTS), IEEE SSCI}, keywords = {isac-www}, pages = {86–93}, publisher = {IEEE}, title = {Space, Time, and Interaction: A Taxonomy of Corner Cases in Trajectory Datasets for Automated Driving}, year = 2022 }
Approximation of the Feasible Parameter Set in Bounded-Error Parameter Estimation of Takagi-Sugeno Fuzzy Models for Large Problems by Using a Ray Shooting Method. Wittich, Felix; Kroll, Andras. In Proceedings of the IEEE International Conference on Fuzzy Systems (FUZZ-IEEE 2022). Padua, Italy, 2022.
- [ BibTeX ]
@inproceedings{Wittich_WCCI_2022, address = {Padua, Italy}, author = {Wittich, Felix and Kroll, Andras}, booktitle = {Proceedings of the IEEE International Conference on Fuzzy Systems (FUZZ-IEEE 2022)}, keywords = {isac-www}, month = {18.-23. July}, title = {Approximation of the Feasible Parameter Set in Bounded-Error Parameter Estimation of Takagi-Sugeno Fuzzy Models for Large Problems by Using a Ray Shooting Method}, year = 2022 }
Method and Experimental Investigation of Surface Heat Dissipation Measurement using 3D Thermography. Schmoll, Robert; Schramm, Sebastian; Breitenstein, Tom; Kroll, Andreas. In Journal of Sensors and Sensor Systems (JSSS), 11, bll 41–49. 2022.
- [ BibTeX ]
- [ URL ]
@article{SchmollJSSS2021, author = {Schmoll, Robert and Schramm, Sebastian and Breitenstein, Tom and Kroll, Andreas}, journal = {Journal of Sensors and Sensor Systems (JSSS)}, keywords = {isac-www}, pages = {41–49}, title = {Method and Experimental Investigation of Surface Heat Dissipation Measurement using 3D Thermography}, volume = 11, year = 2022 }
Predicting the compressive strength of concrete up to 28 days-ahead: Comparison of machine learning algorithms on benchmark datasets. Rezazadeh, Farzad; Kroll, Andreas. In 32. Workshop Computational Intelligence, bll 53–75. KIT Scientific Publishing, Berlin, 2022.
- [ BibTeX ]
- [ URL ]
@inproceedings{RezazadehGMACI2022, author = {Rezazadeh, Farzad and Kroll, Andreas}, booktitle = {32. Workshop Computational Intelligence}, keywords = {isac-www}, month = {1.-2. December}, organization = {GMA-FA 5.14}, pages = {53-75}, publisher = {KIT Scientific Publishing}, title = {Predicting the compressive strength of concrete up to 28 days-ahead: Comparison of machine learning algorithms on benchmark datasets}, year = 2022 }
Investigation of processing windows in additive manufacturing of AlSi10Mg for faster production utilizing data-driven modelling. Engelhardt, A; Kahl, M; Richter, J; Krooß, P; Kroll, A; Niendorf, T. In Additive Manufacturing, 55. Elsevier, 2022.
- [ BibTeX ]
- [ URL ]
@article{ADDMA2021, author = {Engelhardt, A and Kahl, M and Richter, J and Krooß, P and Kroll, A and Niendorf, T}, journal = {Additive Manufacturing}, keywords = {isac-www}, note = 102858, publisher = {Elsevier}, title = {Investigation of processing windows in additive manufacturing of AlSi10Mg for faster production utilizing data-driven modelling}, volume = 55, year = 2022 }
An internal dynamics approach to predicting batch-end product quality in plastic injection molding using Recurrent Neural Networks. Rehmer, Alexander; Klute, Marco; Kroll, Andreas; Heim, Hans-Peter. In IFAC-PapersOnLine, Vol. 53, bll 1427–1432. Elsevier, Trieste, Italy, 2022.
- [ Abstract ]
- [ BibTeX ]
Recurrent Neural Networks are applied in areas such as speech recognition, natural language and video processing, and the identification of nonlinear state space models. Conventional Recurrent Neural Networks, e.g. the Elman Network, are hard to train. A more recently developed class of recurrent neural networks, so-called Gated Units, outperform their counterparts on virtually every task. This paper aims to provide additional insights into the differences between RNNs and Gated Units in order to explain the superior perfomance of gated recurrent units. It is argued, that Gated Units are easier to optimize not because they solve the vanishing gradient problem, but because they circumvent the emergence of large local gradients.
@inproceedings{Rehmer-CCTA-2022, abstract = {Recurrent Neural Networks are applied in areas such as speech recognition, natural language and video processing, and the identification of nonlinear state space models. Conventional Recurrent Neural Networks, e.g. the Elman Network, are hard to train. A more recently developed class of recurrent neural networks, so-called Gated Units, outperform their counterparts on virtually every task. This paper aims to provide additional insights into the differences between RNNs and Gated Units in order to explain the superior perfomance of gated recurrent units. It is argued, that Gated Units are easier to optimize not because they solve the vanishing gradient problem, but because they circumvent the emergence of large local gradients.}, address = {Trieste, Italy}, author = {Rehmer, Alexander and Klute, Marco and Kroll, Andreas and Heim, Hans-Peter}, booktitle = {6th IEEE Conference on Control Technology and Applications (CCTA)}, journal = {IFAC-PapersOnLine}, keywords = {isac-www}, month = {22.-25- August}, organization = {IFAC}, pages = {1427-1432}, publisher = {Elsevier}, title = {An internal dynamics approach to predicting batch-end product quality in plastic injection molding using Recurrent Neural Networks}, volume = 53, year = 2022 }
A Practical Evaluation of Active Learning Approaches for Object Detection. Schneegans, Jan; Bieshaar, Maarten; Sick, Bernhard. In Workshop on Interactive Adaptive Learning (IAL), ECML PKDD, bll 49–67. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Deep learning models are trained most successfully in a supervised manner, which requires vast amounts of annotated data. With active learning, the annotation process can be made much more efficient by intelligently selecting the most valuable batches of samples to annotate and train on. Those samples are selected based on their utility regarding the training algorithm. In this work, we examine a wide range of such selection criteria for the task of object detection as performed by the widely applied Faster R-CNN model. We focus on the large and diverse BDD100K autonomous driving dataset, paying special attention to the model’s sub-tasks and the dataset’s meta information. Furthermore, we distinguish between approaches that select samples based on aleatoric or epistemic uncertainty. A selection of evaluation measures that cover specific error sources and the overall model performance suggests that there is little difference between the individual active learning approaches, even in regards to their specialized focus on different model parts and sub-tasks. We conclude with a detailed discussion of the implied mechanisms regarding the active learning approaches that seem to affect model performances.
@inproceedings{schneegans2022practical, abstract = {Deep learning models are trained most successfully in a supervised manner, which requires vast amounts of annotated data. With active learning, the annotation process can be made much more efficient by intelligently selecting the most valuable batches of samples to annotate and train on. Those samples are selected based on their utility regarding the training algorithm. In this work, we examine a wide range of such selection criteria for the task of object detection as performed by the widely applied Faster R-CNN model. We focus on the large and diverse BDD100K autonomous driving dataset, paying special attention to the model’s sub-tasks and the dataset’s meta information. Furthermore, we distinguish between approaches that select samples based on aleatoric or epistemic uncertainty. A selection of evaluation measures that cover specific error sources and the overall model performance suggests that there is little difference between the individual active learning approaches, even in regards to their specialized focus on different model parts and sub-tasks. We conclude with a detailed discussion of the implied mechanisms regarding the active learning approaches that seem to affect model performances.}, author = {Schneegans, Jan and Bieshaar, Maarten and Sick, Bernhard}, booktitle = {Workshop on Interactive Adaptive Learning (IAL), ECML PKDD}, keywords = {isac-www}, pages = {49–67}, title = {A Practical Evaluation of Active Learning Approaches for Object Detection}, year = 2022 }
Eine Python-Toolbox zur datengetriebenen Modellierung des Spritzgieprozsses und Lösung von Optimalsteuerungsproblemen zur Steuerung der Bauteilqualität. Rehmer, Alexander; Kroll, Andreas. In 32. Workshop Computational Intelligence, bll 133–150. KIT Scientific Publishing, Berlin, 2022.
- [ BibTeX ]
- [ URL ]
@inproceedings{RehmerGMACI2022, address = {Berlin}, author = {Rehmer, Alexander and Kroll, Andreas}, booktitle = {32. Workshop Computational Intelligence}, keywords = {isac-www}, month = {1. - 2. Dezember 2022}, organization = {GMA-FA 5.14}, pages = {133-150}, publisher = {KIT Scientific Publishing}, title = {Eine Python-Toolbox zur datengetriebenen Modellierung des Spritzgieprozsses und Lösung von Optimalsteuerungsproblemen zur Steuerung der Bauteilqualität}, year = 2022 }
On affine quasi-LPV System Identification with unknown state-scheduling using (deep) Recurrent Neural Networks. Rehmer, Alexander; Kroll, Andreas. In IFAC-PapersOnLine, bll 446–451. Sinaia, Romania, 2022.
- [ Abstract ]
- [ BibTeX ]
Recurrent Neural Networks are applied in areas such as speech recognition, natural language and video processing, and the identification of nonlinear state space models. Conventional Recurrent Neural Networks, e.g. the Elman Network, are hard to train. A more recently developed class of recurrent neural networks, so-called Gated Units, outperform their counterparts on virtually every task. This paper aims to provide additional insights into the differences between RNNs and Gated Units in order to explain the superior perfomance of gated recurrent units. It is argued, that Gated Units are easier to optimize not because they solve the vanishing gradient problem, but because they circumvent the emergence of large local gradients.
@inproceedings{Rehmer-ICSTCC-2022, abstract = {Recurrent Neural Networks are applied in areas such as speech recognition, natural language and video processing, and the identification of nonlinear state space models. Conventional Recurrent Neural Networks, e.g. the Elman Network, are hard to train. A more recently developed class of recurrent neural networks, so-called Gated Units, outperform their counterparts on virtually every task. This paper aims to provide additional insights into the differences between RNNs and Gated Units in order to explain the superior perfomance of gated recurrent units. It is argued, that Gated Units are easier to optimize not because they solve the vanishing gradient problem, but because they circumvent the emergence of large local gradients.}, address = {Sinaia, Romania}, author = {Rehmer, Alexander and Kroll, Andreas}, booktitle = {Proceedings of the 26th International Conference on System Theory, Control and Computing (ICSTCC)}, journal = {IFAC-PapersOnLine}, keywords = {isac-www}, month = {19.-21- October}, pages = {446-451}, title = {On affine quasi-LPV System Identification with unknown state-scheduling using (deep) Recurrent Neural Networks}, year = 2022 }
Calibration and validation of micromagnetic data for non-destructive analysis of near-surface properties after hard turning. Wegener, Thomas; Liehr, Alexander; Bolender, Artjom; Degener, Sebastian; Wittich, Felix; Kroll, Andreas; Niendorf, Thomas. In HTM Journal of Heat Treatment and Materials, 77(2), bll 156–172. 2022.
- [ BibTeX ]
@article{Wegener_HTM2021, author = {Wegener, Thomas and Liehr, Alexander and Bolender, Artjom and Degener, Sebastian and Wittich, Felix and Kroll, Andreas and Niendorf, Thomas}, journal = {HTM Journal of Heat Treatment and Materials}, keywords = {isac-www}, number = 2, pages = {156 – 172}, title = {Calibration and validation of micromagnetic data for non-destructive analysis of near-surface properties after hard turning}, volume = 77, year = 2022 }
LwTool: A data processing toolkit for building a real-time pressure mapping smart textile software system. Guo, Tao; Huang, Zhixin; Cheng, Jingyuan. In Pervasive and Mobile Computing, 80, bl 101540. Elsevier, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Pressure mapping smart textile is a new type of sensing modality that transforms the pressure distribution over surfaces into digital ”image” and ”video”, that has rich application scenarios in Human Activity Recognition (HAR), because all human activities are linked with force change over certain surfaces. To speed up its application exploration, we propose a toolkit named LwTool for the data processing, including: (a) a feature library, including 1830 ready-to-use temporal and spatial features, (b) a hierarchical feature selection framework that automatically picks out the best features for a new application from the feature library. As real-time processing capability is important for instant user feedback, we emphasize not only on good recognition result but also on reducing time cost when selecting features. Our library and algorithms are validated on Smart-Toy and Smart-Bedsheet applications, an 89.7% accuracy for Smart-Toy and an 83.8% accuracy for Smart-Bedsheet can be achieved (10-fold cross-validation) using our feature library. Adopting the feature selection algorithm, the processing speed is increased by more than 3 times while maintaining high accuracy for both two applications. We believe our method could be a general and powerful toolkit in building real-time recognition software systems for pressure mapping smart textile.
@article{guo2022lwtool, abstract = {Pressure mapping smart textile is a new type of sensing modality that transforms the pressure distribution over surfaces into digital ”image” and ”video”, that has rich application scenarios in Human Activity Recognition (HAR), because all human activities are linked with force change over certain surfaces. To speed up its application exploration, we propose a toolkit named LwTool for the data processing, including: (a) a feature library, including 1830 ready-to-use temporal and spatial features, (b) a hierarchical feature selection framework that automatically picks out the best features for a new application from the feature library. As real-time processing capability is important for instant user feedback, we emphasize not only on good recognition result but also on reducing time cost when selecting features. Our library and algorithms are validated on Smart-Toy and Smart-Bedsheet applications, an 89.7% accuracy for Smart-Toy and an 83.8% accuracy for Smart-Bedsheet can be achieved (10-fold cross-validation) using our feature library. Adopting the feature selection algorithm, the processing speed is increased by more than 3 times while maintaining high accuracy for both two applications. We believe our method could be a general and powerful toolkit in building real-time recognition software systems for pressure mapping smart textile.}, author = {Guo, Tao and Huang, Zhixin and Cheng, Jingyuan}, journal = {Pervasive and Mobile Computing}, keywords = {MachineLearning}, pages = 101540, publisher = {Elsevier}, title = {LwTool: A data processing toolkit for building a real-time pressure mapping smart textile software system}, volume = 80, year = 2022 }
Multi-Task Autoencoders and Transfer Learning for Day-Ahead Wind and Photovoltaic Power Forecasts. Schreiber, Jens; Sick, Bernhard. In Energies, 15(21), bl 8062. MDPI, 2022.
- [ Abstract ]
- [ BibTeX ]
Integrating new renewable energy resources requires robust and reliable forecasts to ensure a stable electrical grid and avoid blackouts. Sophisticated representation learning techniques, such as autoencoders, play an essential role, as they allow for the extraction of latent features to forecast the expected generated wind and photovoltaic power for the next seconds up to days. Thereby, autoencoders reduce the required training time and the time spent in manual feature engineering and often improve the forecast error. However, most current renewable energy forecasting research on autoencoders focuses on smaller forecast horizons for the following seconds and hours based on meteorological measurements. At the same time, larger forecast horizons, such as day-ahead power forecasts based on numerical weather predictions, are crucial for planning loads and demands within the electrical grid to prevent power failures. There is little evidence on the ability of autoencoders and their respective forecasting models to improve through multi-task learning and time series autoencoders for day-ahead power forecasts. We can close these gaps by proposing a multi-task learning autoencoder based on the recently introduced temporal convolution network. This approach reduces the number of trainable parameters by 38 for photovoltaic data and 202 for wind data while having the best reconstruction error compared to nine other representation learning techniques. At the same time, this model decreases the day-ahead forecast error up to 18.3% for photovoltaic parks and 1.5% for wind parks. We round off these results by analyzing the influences of the latent size and the number of layers to fine-tune the encoder for wind and photovoltaic power forecasts.
@article{schreiber2022multi, abstract = {Integrating new renewable energy resources requires robust and reliable forecasts to ensure a stable electrical grid and avoid blackouts. Sophisticated representation learning techniques, such as autoencoders, play an essential role, as they allow for the extraction of latent features to forecast the expected generated wind and photovoltaic power for the next seconds up to days. Thereby, autoencoders reduce the required training time and the time spent in manual feature engineering and often improve the forecast error. However, most current renewable energy forecasting research on autoencoders focuses on smaller forecast horizons for the following seconds and hours based on meteorological measurements. At the same time, larger forecast horizons, such as day-ahead power forecasts based on numerical weather predictions, are crucial for planning loads and demands within the electrical grid to prevent power failures. There is little evidence on the ability of autoencoders and their respective forecasting models to improve through multi-task learning and time series autoencoders for day-ahead power forecasts. We can close these gaps by proposing a multi-task learning autoencoder based on the recently introduced temporal convolution network. This approach reduces the number of trainable parameters by 38 for photovoltaic data and 202 for wind data while having the best reconstruction error compared to nine other representation learning techniques. At the same time, this model decreases the day-ahead forecast error up to 18.3% for photovoltaic parks and 1.5% for wind parks. We round off these results by analyzing the influences of the latent size and the number of layers to fine-tune the encoder for wind and photovoltaic power forecasts.}, author = {Schreiber, Jens and Sick, Bernhard}, journal = {Energies}, keywords = {isac-www}, number = 21, pages = 8062, publisher = {MDPI}, title = {Multi-Task Autoencoders and Transfer Learning for Day-Ahead Wind and Photovoltaic Power Forecasts}, volume = 15, year = 2022 }
On the Extension of the Weisfeiler-Lehman Hierarchy by WL Tests for Arbitrary Graphs. Beddar-Wiesing, Silvia; D’Inverno, Giuseppe Alessio; Graziani, Caterina; Lachi, Veronica; Moallemy-Oureh, Alice; Scarselli, Franco. In Workshp on Mining and Learning on Graphs (MLG), ECML PKDD, bll 1–13. 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Graph isomorphism (GI) has occupied both theoreticians and applied scientists since the early 1950s. Over the years, several approaches and algorithms with which an isomorphism between two graphs can be tested have been developed. A general approach is the Weisfeiler–Lehman (WL) test, which is based on a coloring algorithm and provides a necessary criterion for graph isomorphism. However, the WL test is restricted to examining graphs with only node attributes. Therefore, this paper presents two extensions of the WL algorithm to allow for testing on arbitrary graphs. One considers edge attributes, and the other tests the isomorphism on dynamic graphs. Additionally, we extend the WL-hierarchy by the attributed and dynamic WL tests and show that it is a partial order, which induces a lattice. In the future, this may allow for practical implications coming from lattice theory.
@inproceedings{beddarwiesing2022extension, abstract = {Graph isomorphism (GI) has occupied both theoreticians and applied scientists since the early 1950s. Over the years, several approaches and algorithms with which an isomorphism between two graphs can be tested have been developed. A general approach is the Weisfeiler–Lehman (WL) test, which is based on a coloring algorithm and provides a necessary criterion for graph isomorphism. However, the WL test is restricted to examining graphs with only node attributes. Therefore, this paper presents two extensions of the WL algorithm to allow for testing on arbitrary graphs. One considers edge attributes, and the other tests the isomorphism on dynamic graphs. Additionally, we extend the WL-hierarchy by the attributed and dynamic WL tests and show that it is a partial order, which induces a lattice. In the future, this may allow for practical implications coming from lattice theory.}, author = {Beddar-Wiesing, Silvia and D'Inverno, Giuseppe Alessio and Graziani, Caterina and Lachi, Veronica and Moallemy-Oureh, Alice and Scarselli, Franco}, booktitle = {Workshp on Mining and Learning on Graphs (MLG), ECML PKDD}, keywords = {isac-www}, pages = {1–13}, title = {On the Extension of the {W}eisfeiler-{L}ehman Hierarchy by WL Tests for Arbitrary Graphs}, year = 2022 }
Social Machines. Draude, Claude; Gruhl, Christian; Hornung, Gerrit; Kropf, Jonathan; Lamla, Jörn; Leimeister, Jan Marco; Sick, Bernhard; Stumme, Gerd. In Informatik Spektrum, 45(1), bll 38–42. Springer, 2022.
- [ Abstract ]
- [ BibTeX ]
Social Machines sind ein Paradigma für die Gestaltung soziotechnischer Systeme, die unter Verwendung von Web- und Plattformlösungen das Potenzial digitaler Technologien mit der Eigenlogik sozialer Interaktion, Organisation und Struk- turbildung auf neue Weise zusammenführen. Im Folgenden diskutieren wir das Paradigma der Social Machine aus den Perspektiven der Informatik, der Wirtschaftsinformatik, der Soziologie und des Rechts, um Orientierungspunkte für seine Gestaltung zu identifizieren. Der Begriff ist in der Literatur jedoch bisher nicht abschließend definiert sondern nur durch Beispiele illustriert. In diesem Artikel stellen wir zunächst die folgende Definition zur Diskussion: Social Machines sind soziotechnische Sys- teme, in denen die Prozesse sozialer Interaktion hybrid zwischen menschlichen und maschinellen Akteuren ablaufen und teilweise algorithmisiert sind. Im Anschluss beleuchten wir drei aktuelle, sich gegenseitig bedingende Entwicklungen von Social Machines: die immer stärkere Verschmelzung von Sozialität und Maschine, die Vermessung von Nutzeraktivitäten als Grundstoff gesellschaftlichen Zusammenhalts und die zunehmende Algorithmisierung gesellschaftlicher Prozesse. Ab- schließend diskutieren wir, dass eine teilhabeorientierte, demokratischen Werten folgende Gestaltung von Social Machines die Perspektiven der Nutzungsakzeptanz, der gesellschaftlichen Akzeptabilität und der nachhaltigen Wirtschaftlichkeit adressieren und umsetzen muss.
@article{draude2022social, abstract = {Social Machines sind ein Paradigma für die Gestaltung soziotechnischer Systeme, die unter Verwendung von Web- und Plattformlösungen das Potenzial digitaler Technologien mit der Eigenlogik sozialer Interaktion, Organisation und Struk- turbildung auf neue Weise zusammenführen. Im Folgenden diskutieren wir das Paradigma der Social Machine aus den Perspektiven der Informatik, der Wirtschaftsinformatik, der Soziologie und des Rechts, um Orientierungspunkte für seine Gestaltung zu identifizieren. Der Begriff ist in der Literatur jedoch bisher nicht abschließend definiert sondern nur durch Beispiele illustriert. In diesem Artikel stellen wir zunächst die folgende Definition zur Diskussion: Social Machines sind soziotechnische Sys- teme, in denen die Prozesse sozialer Interaktion hybrid zwischen menschlichen und maschinellen Akteuren ablaufen und teilweise algorithmisiert sind. Im Anschluss beleuchten wir drei aktuelle, sich gegenseitig bedingende Entwicklungen von Social Machines: die immer stärkere Verschmelzung von Sozialität und Maschine, die Vermessung von Nutzeraktivitäten als Grundstoff gesellschaftlichen Zusammenhalts und die zunehmende Algorithmisierung gesellschaftlicher Prozesse. Ab- schließend diskutieren wir, dass eine teilhabeorientierte, demokratischen Werten folgende Gestaltung von Social Machines die Perspektiven der Nutzungsakzeptanz, der gesellschaftlichen Akzeptabilität und der nachhaltigen Wirtschaftlichkeit adressieren und umsetzen muss.}, author = {Draude, Claude and Gruhl, Christian and Hornung, Gerrit and Kropf, Jonathan and Lamla, Jörn and Leimeister, Jan Marco and Sick, Bernhard and Stumme, Gerd}, journal = {Informatik Spektrum}, keywords = {social-machine}, number = 1, pages = {38–42}, publisher = {Springer}, title = {Social Machines}, volume = 45, year = 2022 }
NDNET: A Unified Framework for Anomaly and Novelty Detection. Decke, Jens; Schmeißing, Jörn; Botache, Diego; Bieshaar, Maarten; Sick, Bernhard; Gruhl, Christian. In International Conference on Architecture of Computing Systems (ARCS), bll 197–210. Springer, 2022.
- [ Abstract ]
- [ BibTeX ]
We introduce NDNET (https://novelty-detection.net/p/ndnet), an anomaly and novelty detection library that implements various detection algorithms adjusted for online processing of data streams. The intention of this library is threefold: 1) Make experimentation with different anomaly and novelty detection algorithms simple. 2) Support the development of new novelty detection approaches by providing the mCANDIES framework. 3) Provide fundamentals to analyze and evaluate novelty detection algorithms on data streams. The library is freely available and developed as open-source software.
@inproceedings{decke2022ndnet, abstract = {We introduce NDNET (https://novelty-detection.net/p/ndnet), an anomaly and novelty detection library that implements various detection algorithms adjusted for online processing of data streams. The intention of this library is threefold: 1) Make experimentation with different anomaly and novelty detection algorithms simple. 2) Support the development of new novelty detection approaches by providing the mCANDIES framework. 3) Provide fundamentals to analyze and evaluate novelty detection algorithms on data streams. The library is freely available and developed as open-source software.}, author = {Decke, Jens and Schmeißing, Jörn and Botache, Diego and Bieshaar, Maarten and Sick, Bernhard and Gruhl, Christian}, booktitle = {International Conference on Architecture of Computing Systems (ARCS)}, keywords = {AnomalyDetection}, pages = {197–210}, publisher = {Springer}, title = {{NDNET}: {A} Unified Framework for Anomaly and Novelty Detection}, year = 2022 }
Self-Aware Microsystems. Gruhl, Christian; Tomforde, Sven; Sick, Bernhard. In Workshop on Self-Improving System Integration (SISSY), ACSOS, bll 126–127. IEEE, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Besides infrastructure-based solutions, small-scale microsystems are increasingly connected and interoperating with others in changing environments. Following the same motivation as for self-adaptive systems, this demands novel attempts to counter complexity and provide autonomous decision freedom that paves the path towards resilient and flexible behaviour. In this article, we outline a vision for research on self-aware microsystems (SAM) that aims at providing technical solutions that are especially relevant for initiatives such as self-improving system integration.
@inproceedings{gruhl2022self, abstract = {Besides infrastructure-based solutions, small-scale microsystems are increasingly connected and interoperating with others in changing environments. Following the same motivation as for self-adaptive systems, this demands novel attempts to counter complexity and provide autonomous decision freedom that paves the path towards resilient and flexible behaviour. In this article, we outline a vision for research on self-aware microsystems (SAM) that aims at providing technical solutions that are especially relevant for initiatives such as self-improving system integration.}, author = {Gruhl, Christian and Tomforde, Sven and Sick, Bernhard}, booktitle = {Workshop on Self-Improving System Integration (SISSY), ACSOS}, keywords = {Online}, pages = {126–127}, publisher = {IEEE}, title = {Self-Aware Microsystems}, year = 2022 }
Three-dimensional close-to-substrate trajectories of magnetic microparticles in dynamically changing magnetic field landscapes. Huhnstock, Rico; Reginka, Meike; Sonntag, Claudius; Merkel, Maximilian; Dingel, Kristina; Sick, Bernhard; Vogel, Michael; Ehresmann, Arno. In Scientific Reports, 12(1), bll 1–10. Springer Nature, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
The transport of magnetic particles (MPs) by dynamic magnetic field landscapes (MFLs) using magnetically patterned substrates is promising for the development of Lab-on-a-chip (LOC) systems. The inherent close-to-substrate MP motion is sensitive to changing particle–substrate interactions. Thus, the detection of a modified particle–substrate separation distance caused by surface binding of an analyte is expected to be a promising probe in analytics and diagnostics. Here, we present an essential prerequisite for such an application, namely the label-free quantitative experimental determination of the three-dimensional trajectories of superparamagnetic particles (SPPs) transported by a dynamically changing MFL. The evaluation of defocused SPP images from optical bright-field microscopy revealed a “hopping”-like motion of the magnetic particles, previously predicted by theory, additionally allowing a quantification of maximum jump heights. As our findings pave the way towards precise determination of particle–substrate separations, they bear deep implications for future LOC detection schemes using only optical microscopy.
@article{huhnstock2022three, abstract = {The transport of magnetic particles (MPs) by dynamic magnetic field landscapes (MFLs) using magnetically patterned substrates is promising for the development of Lab-on-a-chip (LOC) systems. The inherent close-to-substrate MP motion is sensitive to changing particle–substrate interactions. Thus, the detection of a modified particle–substrate separation distance caused by surface binding of an analyte is expected to be a promising probe in analytics and diagnostics. Here, we present an essential prerequisite for such an application, namely the label-free quantitative experimental determination of the three-dimensional trajectories of superparamagnetic particles (SPPs) transported by a dynamically changing MFL. The evaluation of defocused SPP images from optical bright-field microscopy revealed a “hopping”-like motion of the magnetic particles, previously predicted by theory, additionally allowing a quantification of maximum jump heights. As our findings pave the way towards precise determination of particle–substrate separations, they bear deep implications for future LOC detection schemes using only optical microscopy.}, author = {Huhnstock, Rico and Reginka, Meike and Sonntag, Claudius and Merkel, Maximilian and Dingel, Kristina and Sick, Bernhard and Vogel, Michael and Ehresmann, Arno}, journal = {Scientific Reports}, keywords = {isac-www}, number = 1, pages = {1–10}, publisher = {Springer Nature}, title = {Three-dimensional close-to-substrate trajectories of magnetic microparticles in dynamically changing magnetic field landscapes}, volume = 12, year = 2022 }
Heterogeneous Multi-Source Deep Adaptive Knowledge-Aware Learning for E-Mobility. Ali, Mohammad Wazed. In IEEE International Conference on Autonomic Computing and Self-Organizing Systems Companion (ACSOS-C), bll 57–60. IEEE, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Electromobility (e-mobility) reduces global fossil fuel-based energy demand and promotes the usage of renewable energy sources. Different types of electric vehicle (EVs) production have increased recently due to their need. However, there is no fixed charging behavior, and demand exists. Uncontrolled EV charging may lead to a harmful impact on the power grid. Besides, most renewable energy sources depend highly on the weather and location. Thus, analyzing, forecasting, and adapting EV charging demand is essential to prevent power outages. Deep Learning (DL) models are a de facto standard technology for energy forecasting in many scenarios. Still, there is a lack of real-world data in the e-mobility domain to train a DL model from scratch. Besides, DL models have explainability issues, which can hinder decision-making. This article proposes a novel framework for addressing data scarcity issues, enabling continuous adaptation to change, and improving explainability.
@inproceedings{ali2022heterogeneous, abstract = {Electromobility (e-mobility) reduces global fossil fuel-based energy demand and promotes the usage of renewable energy sources. Different types of electric vehicle (EVs) production have increased recently due to their need. However, there is no fixed charging behavior, and demand exists. Uncontrolled EV charging may lead to a harmful impact on the power grid. Besides, most renewable energy sources depend highly on the weather and location. Thus, analyzing, forecasting, and adapting EV charging demand is essential to prevent power outages. Deep Learning (DL) models are a de facto standard technology for energy forecasting in many scenarios. Still, there is a lack of real-world data in the e-mobility domain to train a DL model from scratch. Besides, DL models have explainability issues, which can hinder decision-making. This article proposes a novel framework for addressing data scarcity issues, enabling continuous adaptation to change, and improving explainability.}, author = {Ali, Mohammad Wazed}, booktitle = {IEEE International Conference on Autonomic Computing and Self-Organizing Systems Companion (ACSOS-C)}, keywords = {isac-www}, pages = {57–60}, publisher = {IEEE}, title = {Heterogeneous Multi-Source Deep Adaptive Knowledge-Aware Learning for E-Mobility}, year = 2022 }
Receding-Horizon Control of Constrained Switched Systems with Neural Networks as Parametric Function Approximators. Markolf, Lukas; Stursberg, Olaf. In SN Computer Science, 4(62). 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
This work studies receding-horizon control of discrete-time switched linear systems subject to polytopic constraints for the continuous states and inputs. The objective is to approximate the optimal receding-horizon control strategy for cases in which the online computation is intractable due to the necessity of solving mixed-integer quadratic programs in each discrete time instant. The proposed approach builds upon an approximated optimal finite-horizon control law in closed-loop form with guaranteed constraint satisfaction. The paper derives the properties of recursive feasibility and asymptotic stability for the proposed approach. A numerical example is provided for illustration and evaluation of the approach.
@article{markolf2022recedinghorizon, abstract = {This work studies receding-horizon control of discrete-time switched linear systems subject to polytopic constraints for the continuous states and inputs. The objective is to approximate the optimal receding-horizon control strategy for cases in which the online computation is intractable due to the necessity of solving mixed-integer quadratic programs in each discrete time instant. The proposed approach builds upon an approximated optimal finite-horizon control law in closed-loop form with guaranteed constraint satisfaction. The paper derives the properties of recursive feasibility and asymptotic stability for the proposed approach. A numerical example is provided for illustration and evaluation of the approach.}, author = {Markolf, Lukas and Stursberg, Olaf}, journal = {SN Computer Science}, keywords = {isac-www}, number = 62, title = {Receding-Horizon Control of Constrained Switched Systems with Neural Networks as Parametric Function Approximators}, volume = 4, year = 2022 }
Artificial intelligence for online characterization of ultrashort X‑ray free‑electron laser pulses. Dingel, Kristina; Otto, Thorsten; Marder, Lutz; Funke, Lars; Held, Arne; Savio, Sara; Hans, Andreas; Hartmann, Gregor; Meier, David; Viefhaus, Jens; Sick, Bernhard; Ehresmann, Arno; Ilchen, Markus; Helml, Wolfram. In Scientific Reports, 12(1), bll 1–14. Nature Publishing Group, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
X-ray free-electron lasers (XFELs) as the world’s brightest light sources provide ultrashort X-ray pulses with a duration typically in the order of femtoseconds. Recently, they have approached and entered the attosecond regime, which holds new promises for single-molecule imaging and studying nonlinear and ultrafast phenomena such as localized electron dynamics. The technological evolution of XFELs toward well-controllable light sources for precise metrology of ultrafast processes has been, however, hampered by the diagnostic capabilities for characterizing X-ray pulses at the attosecond frontier. In this regard, the spectroscopic technique of photoelectron angular streaking has successfully proven how to non-destructively retrieve the exact time–energy structure of XFEL pulses on a single-shot basis. By using artificial intelligence techniques, in particular convolutional neural networks, we here show how this technique can be leveraged from its proof-of-principle stage toward routine diagnostics even at high-repetition-rate XFELs, thus enhancing and refining their scientific accessibility in all related disciplines.
@article{dingel2022artificial, abstract = {X-ray free-electron lasers (XFELs) as the world’s brightest light sources provide ultrashort X-ray pulses with a duration typically in the order of femtoseconds. Recently, they have approached and entered the attosecond regime, which holds new promises for single-molecule imaging and studying nonlinear and ultrafast phenomena such as localized electron dynamics. The technological evolution of XFELs toward well-controllable light sources for precise metrology of ultrafast processes has been, however, hampered by the diagnostic capabilities for characterizing X-ray pulses at the attosecond frontier. In this regard, the spectroscopic technique of photoelectron angular streaking has successfully proven how to non-destructively retrieve the exact time–energy structure of XFEL pulses on a single-shot basis. By using artificial intelligence techniques, in particular convolutional neural networks, we here show how this technique can be leveraged from its proof-of-principle stage toward routine diagnostics even at high-repetition-rate XFELs, thus enhancing and refining their scientific accessibility in all related disciplines.}, author = {Dingel, Kristina and Otto, Thorsten and Marder, Lutz and Funke, Lars and Held, Arne and Savio, Sara and Hans, Andreas and Hartmann, Gregor and Meier, David and Viefhaus, Jens and Sick, Bernhard and Ehresmann, Arno and Ilchen, Markus and Helml, Wolfram}, journal = {Scientific Reports}, keywords = {isac-www}, number = 1, pages = {1–14}, publisher = {Nature Publishing Group}, title = {Artificial intelligence for online characterization of ultrashort X‑ray free‑electron laser pulses}, volume = 12, year = 2022 }
Proactive hybrid learning and optimisation in self-adaptive systems: The swarm-fleet infrastructure scenario. Krupitzer, Christian; Gruhl, Christian; Sick, Bernhard; Tomforde, Sven. In Information and Software Technology, 145, bl 106826. Elsevier, 2022.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Context: Smart and adaptive Systems, such as self-adaptive and self-organising (SASO) systems, typically consist of a large set of highly autonomous and heterogeneous subsystems that are able to adapt their behaviour to the requirements of ever-changing, dynamic environments. Their successful operation is based on appropriate modelling of the internal and external conditions. Objective: The control problem for establishing a near-to-optimal coordinated behaviour of systems with multiple, potentially conflicting objectives is either approached in a distributed (i.e., fully autonomous by the autonomous subsystems) or in a centralised way (i.e. one instance controlling the optimisation and planning process). In the distributed approach, selfish behaviour and being limited to local knowledge may lead to sub-optimal system behaviour, while the centralised approach ignores the autonomy and the coordination efforts of parts of the system. Method: In this article, we present a concept for a hybrid (i.e., integrating a central optimisation with a distributed decision-making process) system management that combines local reinforcement learning and self-awareness mechanisms of fully autonomous subsystems with external system-wide planning and optimisation of adaptation freedom that steers the behaviour dynamically by issuing plans and guidelines augmented with incentivisation schemes. Results: This work addresses the inherent uncertainty of the dynamic system behaviour, the local autonomous and context-aware learning of subsystems, and proactive control based on adaptiveness. We provide the ‘swarm-fleet infrastructure’—a self-organised taxi service established by autonomous, privately-owned cars—as a testbed for structured comparison of systems. Conclusion: The ‘swarm-fleet infrastructure’ supports the advantages of a proactive hybrid self-adaptive and self-organising system operation. Further, we provide a system model to combine the system-wide optimisation while ensuring local decision-making through reinforcement learning for individualised configurations.
@article{krupitzer2022proactive, abstract = {Context: Smart and adaptive Systems, such as self-adaptive and self-organising (SASO) systems, typically consist of a large set of highly autonomous and heterogeneous subsystems that are able to adapt their behaviour to the requirements of ever-changing, dynamic environments. Their successful operation is based on appropriate modelling of the internal and external conditions. Objective: The control problem for establishing a near-to-optimal coordinated behaviour of systems with multiple, potentially conflicting objectives is either approached in a distributed (i.e., fully autonomous by the autonomous subsystems) or in a centralised way (i.e. one instance controlling the optimisation and planning process). In the distributed approach, selfish behaviour and being limited to local knowledge may lead to sub-optimal system behaviour, while the centralised approach ignores the autonomy and the coordination efforts of parts of the system. Method: In this article, we present a concept for a hybrid (i.e., integrating a central optimisation with a distributed decision-making process) system management that combines local reinforcement learning and self-awareness mechanisms of fully autonomous subsystems with external system-wide planning and optimisation of adaptation freedom that steers the behaviour dynamically by issuing plans and guidelines augmented with incentivisation schemes. Results: This work addresses the inherent uncertainty of the dynamic system behaviour, the local autonomous and context-aware learning of subsystems, and proactive control based on adaptiveness. We provide the ‘swarm-fleet infrastructure’—a self-organised taxi service established by autonomous, privately-owned cars—as a testbed for structured comparison of systems. Conclusion: The ‘swarm-fleet infrastructure’ supports the advantages of a proactive hybrid self-adaptive and self-organising system operation. Further, we provide a system model to combine the system-wide optimisation while ensuring local decision-making through reinforcement learning for individualised configurations.}, author = {Krupitzer, Christian and Gruhl, Christian and Sick, Bernhard and Tomforde, Sven}, journal = {Information and Software Technology}, keywords = {isac-www}, pages = 106826, publisher = {Elsevier}, title = {Proactive hybrid learning and optimisation in self-adaptive systems: The swarm-fleet infrastructure scenario}, volume = 145, year = 2022 }

Publikationen

2023

2022

Aktuelles

Quicklinks

Beteiligte Fachgebiete