Talk: Deep and Shallow Generative Models of Images by Yair Weiss

Title

Deep and Shallow Generative Models of Images

Abstract

Recent years have shown dramatic progress in the visual quality of samples produced by deep generative models. In this talk, I will present work from my lab that asks whether these models have learned good image priors in the traditional sense. Overall, our results suggest that the answer is “no” and that oftentimes shallow models learn better priors than deep models, even though the visual quality of shallow models is much worse.

Bio

Yair Weiss is Professor at the Hebrew University of Jerusalem. Yair Weiss’s research interests include human and machine vision, machine learning, Bayesian methods, and neural computation. You can find his publications here>>.

Yair Weiss is visiting the Pioneer Centre for AI from August 7-18. Contact aicentre@ku.dk, or Yair directly, if you would like to set up an individual meeting.