Application of field theory concepts or techniques to neural networks.

General: Modular model with choosable modules like reversibility, better parallel methods, fractional Fourier transform...

Neural Field bot: more hard coded than General, use of Hamiltonian (generating functional/autograd gradient, canonical transformations, ...) Lagrangian (closer to general but less flexible)

Mainly trained for:

chatbot: just simple text-completion for now (penn_treebank)

stockbot: predict one ohlcv stock based on past data of multiple ohlcv stocks (alpaca)