Understanding Neural Networks with Information Theory

Description

Neural networks are among the most powerful the least understood machine learning architectures. Recently, information theory has been used to open these proverbial "black boxes". This talk starts with a brief, superficial introduction to neural networks before giving an overview of recent information-theoretic approaches to understand and characterize neural networks. Finally, we will investigate how the importance of individual neurons can be quantified with information quantities.

Period	21 Oct 2020
Event title	Lunchtime Lecture
Event type	Other
Degree of Recognition	Local