Lipschitz networks and energy movers distance

by Dr Niklas Nolte (Massachusetts Institute of Technology)

Thursday 15 Dec 2022, 15:30 → 16:30 Europe/Berlin

CP-03-123 (TU Dortmund)

In this talk I will discuss the concept of neural networks with a controllable Lipschitz bound and their applications in High Energy Physics. This work goes over two aspects. Firstly, a provable robustness guarantee and additional benefits for safety-critical applications. Secondly, its use for differentiable estimation of Wasserstein-1 optimal transport potentials for the creation of discriminatory observables, specifically for jet analysis.

slides.pdf