Energy Prediction
The peanut models predict potential energy surfaces based on the learned representations.
All models predict atom-wise energy contributions which are then aggregated to obtain the total energy for the molecular system.
\[E^{(b)} = \sum\limits_{i \in b} E_i.\]
where \(b\) denotes the batch during each training step, and \(E_i\) is the energy contribution from atom \(i\). The energy prediction head typically consists of several fully connected layers that take the final atom-wise representations as input and output scalar energy contributions for each atom.