SeminarPast EventDeep Learning

The Neural Race Reduction: Dynamics of nonlinear representation learning in deep architectures

Schedule
Friday, April 14, 2023
06:30 UTC
Andrew Saxe

Prof.

UCL

Host: NYU Swartz

Event Information

Host

NYU Swartz

Duration

70 minutes

Abstract

What is the relationship between task, network architecture, and population activity in nonlinear deep networks? I will describe the Gated Deep Linear Network framework, which schematizes how pathways of information flow impact learning dynamics within an architecture. Because of the gating, these networks can compute nonlinear functions of their input. We derive an exact reduction and, for certain cases, exact solutions to the dynamics of learning. The reduction takes the form of a neural race with an implicit bias towards shared representations, which then govern the model’s ability to systematically generalize, multi-task, and transfer. We show how appropriate network architectures can help factorize and abstract knowledge. Together, these results begin to shed light on the links between architecture, learning dynamics and network performance.

We use essential cookies to run the site. Analytics cookies are optional and help us improve World Wide. Learn more.