Your Brain on Energy-Based Models - Applying and Scaling EBMs to Problems

Explore the applications and scaling of Energy-Based Models (EBMs) in machine learning through this 45-minute seminar presented by Will Grathwohl from the University of Toronto. Delve into the appeal of generative models and their various applications, examining approximate likelihood gradient and architectures for classification. Gain insights into a different perspective on EBMs and discover results in hybrid modeling, calibration, out-of-distribution detection, and adversarial robustness. Learn how EBMs can be applied to solve problems of interest in the field of machine learning, expanding your understanding of this powerful modeling approach.

Intro
the appeal of generative models
some applications of generative models
approximate likelihood gradient
architectures for classification
a different perspective
results: hybrid modeling
results: calibration
results: out-of-distribution detection
results: adversarial robustness