On Monday, 8th July 2024, Prof. Dr. Felix Dietrich from the chair of Physics-enhanced Machine Learning will talk about current research in his group as part of the TopMath Alumni Speakers Series. The talk will take place at 3 PM in Garching, room 00.10.011. Anyone interested is welcome to attend – for organizational reasons we’d like to ask you to register by email to topmath(at)ma.tum.de by Wednesday, 3 July 3.
For this lecture in the TopMath Alumni Speakers Series, we have invited Prof. Dr. Felix Dietrich. Prof. Dietrich, TopMath class of 2012/13, is one of the TopMath alumni who is continuing his scientific career at his alma mater. Before taking the chair at the Technical University of Munich he was an Emmy Noether Research Group Leader at the Chair of Scientific Computing, he was a Postdoctoral Fellow at the Department of Chemical and Biomolecular Engineering, Whiting School of Engineering at Johns Hopkins University and a Visiting Research Collaborator at Princeton University, where he worked with Prof. Kevrekidis.
He is interested in the analysis and development of numerical algorithms for machine learning. This covers algorithms to enable, accelerate, and optimize simulation and analysis of complex dynamical systems, as well as nonlinear manifold learning techniques, including data-driven approximations of Koopman and Laplace operators.
During the talk, current research in his group regarding randomly sampling parameters of artificial neural networks from certain probability distributions will be discussed. In many cases, this approach outperforms backpropagation-based optimization by several orders of magnitude, for both training speed and accuracy.