HEMI Artificial Intelligence for Materials Seminar: Yannis Kevrekidis and Felix Kemeth

Description

Yannis Kevrekidis and Felix Kemeth will discuss learning (possibly physics informed) predictive laws at the macroscopic level from "big" data at the microscopic level (in this case, computational data that come from fine-scale simulations) for the Hopkins Extreme Materials Institute. One example will involve e-coli swimming, modeled at the individual level, and the subsequent learning of macroscopic chemotactic closures. The second example will involve networks of coupled oscillators (and coupled neurons), and the discovery of "emergent spaces" in which useful collective models can then be learned (using tools like Gaussian Processes, Manifold Learning, or (Deep) Neural Networks).

Please attend the event by using the Microsoft Teams link.

Felix Kemeth is a postdoctoral research at the Department of Chemical and Biomolecular Engineering at Johns Hopkins University. His research focuses on learning dynamical models from data, in particular in the context of multiagent systems. Before joining Johns Hopkins, Felix Kemeth worked as an AI strategist at the Fraunhofer Institute for Integrated Circuits in Erlangen, Germany, where he developed machine learning tools for automated electrocardiogram classification.

Yannis Kevrekidis is a Bloomberg Distinguished Professor in the Departments of Chemical and Biomolecular Engineering and Applied Mathematics and Statistics and in the School of Medicine's Department of Urology. He was an undergraduate engineer in Greece, his graduate studies were in chemical engineering and mathematics in Minnesota, he was a postdoc in Los Alamos when the Soviet Union still existed, and he taught at Princeton for 31 years before joining Johns Hopkins. His research interests have developed from (a) nonlinear dynamics, pattern formation, and their scientific computation to (b) multiscale scientific computing algorithms (he developed the so-called equation-free/variable-free approach to multiscale computation) to (c) manifold learning and data-driven modeling, returning in recent years to nonlinear identification using neural networks that we worked on in the 1990s. Recent interests include data fusion and transformation across different first principles, data driven, and physics-informed models. Applications range from chemical and biochemical/cellular dynamics to materials design and transport.