Tag
This paper introduces MF-Net, a recurrent dynamical model that represents multivariate systems through a shared field state and learns a mechanical transition for joint evolution. It achieves competitive forecasting while enabling interpretable structural readout of learned relations.