Title:	Isoperimetry in deep and reinforcement learning
DNr:	Berzelius-2025-12
Project Type:	LiU Berzelius
Principal Investigator:	Emilio Jorge <emilio.jorge@chalmers.se>
Affiliation:	Chalmers tekniska högskola
Duration:	2025-01-13 – 2025-08-01
Classification:	10210
Keywords:

Abstract

I have studied how the properties of isoperimetry can yield beneficial concentration results in both regular deep learning as well as reinforcement learning. Now I want to experimentally verify if these conditions hold on a set of datasets, but approximately computing the desired constants. This is done to complete the story of my WASP PhD thesis.