Symmetries in PAC-Bayesian Learning

Armin Beck, Peter Ochs

Symmetries are known to improve the empirical performance of machine learning models, yet theoretical guarantees explaining these gains remain limited. Prior work has focused mainl…