Quantum Theory Seminar - Yasaman Bahri (Google Deepmind)

 Friday, February 23rd, 2024 at 3:30pm to 4:30pm

 Clark Hall, 701
Central Campus

Title: Theories of Learning in Deep Neural Networks

Abstract: Recent years have witnessed unprecedented advancements in the development of artificial intelligence built using artificial neural networks. While neural networks are typically treated as black boxes, uncovering and distilling the fundamental principles behind how they work is a challenge for which ideas, tools, and methodology from physics are useful. A central question is to understand how computation in deep neural networks emerges from the collective dynamics and interactions of simple elements. I will describe some highlights from my research taking a first-principles approach to understanding learning in artificial neural networks through the lens of exactly solvable models and mean-field theories, statistical physics, and nonlinear dynamics.

Faculty Hosts: Eun-Ah Kim, Chao-Ming Jian, and Debanjan Chowdhury

Streaming site:

https://www.youtube.com/@quantumtheoryseminar6651/streams