Approximate Equivariance via Projection-based Regularisation

Torben Berndt, Jan Stühmer

Equivariance is a powerful inductive bias in neural networks, improving generalisation and physical consistency. Recently, however, non-equivariant models have regained attention,…