NAISS
SUPR
NAISS Projects
SUPR
Data Augmentation for Equivariant Bayesian Neural Networks
Dnr:

NAISS 2026/4-1491

Type:

NAISS Small

Principal Investigator:

Miaowen Dong

Affiliation:

Chalmers tekniska högskola

Start Date:

2026-08-31

End Date:

2027-09-01

Primary Classification:

10210: Artificial Intelligence

Allocation

Abstract

This project develops Bayesian methods for equivariant neural networks, building on prior work showing that data augmentation can induce group equivariance in variational Bayesian neural networks (arXiv:2606.26273). Empirically validating our theoretical results requires training equivariant network variants under group data augmentation across multiple random seeds, dataset sizes, and symmetry groups, which motivates the GPU allocation requested here. The project is carried out at the Department of Mathematical Sciences, Chalmers University of Technology, under the supervision of Jan Gerken.