Karolina Klockmann (Univ. Kassel): Differential Privacy for Time Series Data

Zoom Link:
https://uni-kassel.zoom.us/j/96217091997?pwd=RVRaVVBRclFJYU9jczNZSWF3SXI2QT09

Meeting ID: 962 1709 1997
Passcode: cauchy

Abstract:

In the era of big data, the protection of personal data is an important topic. However, a high degree of data privatization can reduce the usefulness of statistical methods and the informative value of data analysis. At the same time, traditional privacy methods, such as anonymization, are becoming less effective due to the increasing availability of publicly available information, which makes it easier to link anonymized data with other dataset and re-identify individuals. This highlights the need for more robust privacy frameworks, such as Differential Privacy. By introducing controlled noise to the data, Differential Privacy ensures privacy while still allowing meaningful conclusions to be drawn. From a theoretical perspective, the reduction in the utility of a statistical estimator under Differential Privacy can be quantified by the slowdown in its convergence rate to the true parameter value, when the size of the dataset increases. In this talk, we discuss this trade-off in the context of time series data, specifically focusing on the estimation of the covariance structure.