Hessian-Free Laplace in Bayesian Deep Learning

arXiv:2403.10671v1 Announce Type: cross
Abstract: The Laplace approximation (LA) of the Bayesian posterior is a Gaussian distribution centered at the maximum a posteriori estimate. Its appeal in Bayesian deep learning stems from the ability to quantify uncertainty post-hoc (i.e., after standard network parameter optimization), the ease of sampling from the approximate posterior, and the analytic form of model evidence. However, an important computational bottleneck of LA is the necessary step of calculating and inverting the Hessian matrix of the log …

abstract approximation arxiv bayesian bayesian deep learning cs.lg deep learning distribution free laplace approximation network optimization posterior sampling standard stat.ml type uncertainty