The High Cost of Data Augmentation for Learning Equivariant Models

According to Noether’s theorem, the presence of a continuous symmetry in a Hamiltonian system is equivalent to the existence of a conserved quantity, yet these symmetries are not always explicitly enforced in data-driven models. There remains a debate whether or not enforcing the symmetry into model architecture is the optimal approach, with a competing approach being to target approximate symmetry through data augmentation. In this work, we study two approaches aimed at improving the symmetry properties of such an augmentation scheme: one based on a quadrature rule for the Haar measure on the compact Lie group encoding the continuous symmetry of interest and one based on random sampling of that Haar measure. We demonstrate both theoretically and empirically that the quadrature augmentation leads to exact symmetry preservation in polynomial models, whereas the random augmentation affords only square-root convergence of the symmetrisation error.