Topic: Realtime Learning and Control Theory
Traditional machine learning methods work offline, calculating the
analysis model from a complete data set. A modern alternative is
online learning, where the model is adapted using new data. Typically
online learning requires incremental learning methods. We describe
some ideas of incremental learning. Second, online learning represents
a shift from pure data analysis to control theory. For illustration,
we will give some examples and talk about reinforcement learning.
Michael studied mathematics in Chemnitz und St. Petersburg. He specialized in numerical analysis and received the PhD at the TU Chemnitz. In 1998 he co-founded the prudsys AG, a vendor of realtime analytics for retail. Since 2017 he manages the Signal Cruncher GmbH, a spin-off of prudsys specialized in realtime analytics for IoT.