English

Robustness to corruption in pre-trained Bayesian neural networks

Machine Learning 2023-02-24 v3 Artificial Intelligence

Abstract

We develop ShiftMatch, a new training-data-dependent likelihood for robustness to corruption in Bayesian neural networks (BNNs). ShiftMatch is inspired by the training-data-dependent "EmpCov" priors from Izmailov et al. (2021a), and efficiently matches test-time spatial correlations to those at training time. Critically, ShiftMatch is designed to leave the neural network's training time likelihood unchanged, allowing it to use publicly available samples from pre-trained BNNs. Using pre-trained HMC samples, ShiftMatch gives strong performance improvements on CIFAR-10-C, outperforms EmpCov priors (though ShiftMatch uses extra information from a minibatch of corrupted test points), and is perhaps the first Bayesian method capable of convincingly outperforming plain deep ensembles.

Keywords

Cite

@article{arxiv.2206.12361,
  title  = {Robustness to corruption in pre-trained Bayesian neural networks},
  author = {Xi Wang and Laurence Aitchison},
  journal= {arXiv preprint arXiv:2206.12361},
  year   = {2023}
}

Comments

Published in the International Conference on Learning Representations (ICLR) 2023

R2 v1 2026-06-24T12:03:15.831Z