Symmetry-Aware Generative Modeling through Learned Canonicalization
This addresses a problem in AI for science applications like drug discovery and physics simulations, offering an incremental improvement over existing methods.
The paper tackles generative modeling of symmetric densities by proposing a learned canonicalization approach that maps samples to a canonical pose, enabling the use of non-equivariant generative models. Preliminary results on molecular modeling show improved sample quality and faster inference time.
Generative modeling of symmetric densities has a range of applications in AI for science, from drug discovery to physics simulations. The existing generative modeling paradigm for invariant densities combines an invariant prior with an equivariant generative process. However, we observe that this technique is not necessary and has several drawbacks resulting from the limitations of equivariant networks. Instead, we propose to model a learned slice of the density so that only one representative element per orbit is learned. To accomplish this, we learn a group-equivariant canonicalization network that maps training samples to a canonical pose and train a non-equivariant generative model over these canonicalized samples. We implement this idea in the context of diffusion models. Our preliminary experimental results on molecular modeling are promising, demonstrating improved sample quality and faster inference time.