Permutation invariant polynomial neural network approach to fitting potential energy surfaces | Synapse