Maurizio Filippone and Ba-Hien Tran
Associate Professor and PhD Student, EURECOM
- Date & time
- Monday, May 2, 2022 · 1:00 PM
- Location
- Donald Bren Hall 4011
Abstract
The Bayesian treatment of neural networks dictates that a prior distribution is specified over their weight and bias parameters, which is challenging because modern networks have huge numbers of parameters and non-linearities. Gaussian processes instead offer a rigorous non-parametric framework to define priors over the space of functions. In this talk we introduce a robust framework to impose functional priors on modern neural networks by minimizing the Wasserstein distance between samples of stochastic processes, and extend it to model selection for Bayesian autoencoders. Coupling these priors with scalable MCMC offers systematically large performance improvements over alternative choices of priors.
About the speaker
Maurizio Filippone is AXA Chair of Computational Statistics and Associate Professor at EURECOM, Sophia Antipolis, France; his research develops tractable and scalable Bayesian inference for Gaussian processes and deep networks. Ba-Hien Tran is a PhD student at EURECOM advised by Prof. Filippone, focusing on accelerating inference for deep probabilistic modeling.