Atom-centered neural networks represent the state-of-the-art for approximating quantum chemical properties of molecules, such as internal energies, but the final atom pooling operation that is necessary to convert from atomic to molecular representations in most models remains relatively undeveloped. Here, the authors report a learnable pooling operation, usable as a drop-in replacement, that leverages an attention mechanism to model interactions between atom representations.
- David Buterez
- Jon Paul Janet
- Pietro Liò
