Learning Probabilistic Symmetrization for Architecture Agnostic Equivariance

We present a novel framework to overcome the limitations of equivariant architectures in learning functions with group symmetries.