GraphNeuralNetworks.jl support equivariant neural networks

support equivariant neural networks

Open CarloLucibello opened this issue 2 years ago • 6 comments

[ ] General forms using spherical harmonics https://www.nature.com/articles/s41467-022-29939-5 https://docs.e3nn.org/en/latest/guide/convolution.html
[x] Simpler equivariant layers https://proceedings.mlr.press/v139/satorras21a.html https://github.com/lucidrains/egnn-pytorch https://docs.dgl.ai/en/0.9.x/generated/dgl.nn.pytorch.conv.EGNNConv.html

Nov 08 '22 07:11 CarloLucibello

I was working with e3nn and nequip this summer and would love to see it in Julia. Something like e3nn.jl sounds exciting.

Nov 22 '22 09:11 Dsantra92

have you compared them to the simpler EGNN for your tasks?

Nov 22 '22 10:11 CarloLucibello

By comparison, do you mean performance or implementation wise?

Nov 22 '22 10:11 Dsantra92

Test accuracy. Implementation wise they are very different, e3nn much harder to implement

Nov 22 '22 10:11 CarloLucibello

No, not yet. We were mainly focused on spherical convolutions.

Nov 22 '22 10:11 Dsantra92

Started writing this e3nn.jl library.

Dec 29 '22 01:12 Dsantra92