Stream-based active learning identifies data points within a data stream that can be useful for training the currently used classifier. These data points are unlabeled, so they are passed to an oracle, which assigns (annotates) a class to each of these data points. In the current state-of-the-art, the assumption is that the oracle annotates this data immediately and that this annotation is available immediately. However, due to limited number of experts or complex computer simulations, delays can occur (verification latency). In "Stream-Based Active Learning for Sliding Windows Under Verification Latency", Tuan Pham, Daniel Kottke, Georg Krempl, and Bernhard Sick investigate the impact of verification latencies in stream-based active learning. The authors show that by early forgetting and simulating annotations that are not yet available, state-of-the-art active learning strategies can make a more intelligent selection of data points.