BARC talk by Christoph Hertrich

Wednesday, April 23, 2025, Christoph Hertrich, tenure-track Professor at the University of Technology Nuremberg, Germany, will give a talk on "Understanding Neural Networks via Polyhedral Geometry ".

Abstract:
Neural networks with rectified linear unit (ReLU) activations are one of the standard models in modern machine learning. Despite their practical importance, fundamental theoretical questions concerning ReLU networks remain open until today. For instance, what is the precise set of (piecewise linear) functions representable by ReLU networks with a given depth? And what functions can we represent with polynomial-size neural networks? In this talk Christoph will explain how we can use techniques from polyhedral geometry and combinatorial optimization to make progress towards resolving these questions. 

Bio:
Christoph is a tenure-track Professor for Applied Discrete Mathematics at University of Technology Nuremberg. His research interests span various topics across discrete mathematics, theoretical computer science, and machine learning, with an emphasis on applying techniques from polyhedral geometry and combinatorial optimization to neural network theory.

Previously, in 2023/24, he was a postdoc at Université libre de Bruxelles advised by Samuel Fiorini and partially funded through Christoph's own Marie Skłodowska-Curie fellowship. He paused his stay in Brussels when he acted as a substitute professor for discrete mathematics at Goethe-Universität Frankfurt in the winter semester 2023/24. Furthermore, in 2022/23, Christoph was a postdoc with László Végh at LSE in London. He completed his PhD with Martin Skutella at TU Berlin (2018-22), and his BSc and MSc with Sven O. Krumke at TU Kaiserslautern (2013-18).

Host: 
Amir Yehudayoff