Christoph Hertrich (TU Nürnberg): Understanding Neural Network Expressivity via Polyhedral Geometry

Zoom Link: 
https://uni-kassel.zoom-x.de/j/62114766919?pwd=TwdVfoGuQPbTrHij2KP2WP5DubXa4x.1

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 (pie-
cewise linear) functions representable by ReLU networks with a given depth? And what functions
can we represent with polynomial-size neural networks?
In this talk I will report about recent progress towards resolving such questions using techniques
from polyhedral geometry and combinatorial optimization.

Before this lecture, starting at 4:45 p.m., there will be coffee and tea in Room 1404.

Everyone is warmly invited.

Signed: Prof. Dr. Torsten Mütze