PointNet for WF optimization

[NicoRenaud]

We could try to use PointNet architecture (i.e. deep NN on point cloud : https://github.com/charlesq34/pointnet) to optimize the parameter of the wavefunction.

The sampled points of the density could be used as input of the network which could predict the parameters of the wf (eg the CI coefficients). The loss will be computed from these parameters through the total energy or the variance

That should work.

[NicoRenaud]

We also have a pyTorch implementation done by Lukas for DeepRank: https://github.com/DeepRank/PoNDeR
We could tweak it for pyCHAMP