Implement and train a point cloud autoencoder.

[InfluenceFunctional]

Input point cloud with categorical features (or perhaps other features, e.g., atom types).

Output identical point cloud.

Should be equivariant w.r.t. input cloud.

To-do:

• implement equivariant encoder & decoder
• select a point cloud reconstruction loss

[InfluenceFunctional]

Finished in equivariant branch.