(Function-space) Laplace Approximation for Bayesian Neural Networks

Abstract

In this talk, I’ll present an overview of the Laplace approximation for quantifying uncertainty in Bayesian neural networks. I’ll then introduce our work, name Sparse Function-space Representation (SFR), which can be viewed as a function-space Laplace approximation for BNNs. I’ll demonstrate the proposed approach for quantifying uncertainty in supervised learning, maintaining an expressive functional representation for continual learning, and guiding exploration in model-based reinforcement learning.