A new perspective on building efficient and expressive 3D equivariant graph neural networks

Neural Information Processing Systems 

Geometric deep learning enables the encoding of physical symmetries in modeling 3D objects.