Collected sources and patterns will appear here. Add from search, explore, or the patterns library.
Graph<EquivariantFeatures> -> Graph<EquivariantFeatures>
Compute scalar attention coefficients from coordinate-invariant features to scale and aggregate directional, equivariant coordinate messages.
Problem it solves
Direct transformer self-attention over raw 3D coordinate-based representations violates SE(3) covariance constraints.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.