Simplifying and Generalising Equivariant Geometric Algebra Networks
Yuxin Yao -- University of Cambridge
In this paper we present and explain geometric algebra equivariant neural network mappings, and use these to construct a generalised equivariant geometric algebra transformer very similar to that in Geometric Algebra Transformers by Brehmer et al. We then present tests conducted on an n-body dynamics problem and a protein structure prediction problem, using a conformal geometric algebra (CGA) implementation of the transformer blocks.