On architectural choices for interpretability and thermodynamic consistency in Physically Recurrent Neural Networks in the low-data regime
An adjoint encoder-decoder structure with positive scalar contributions ensures energy consistency across scales and non-negative dissipation, leading to even lower training requirements, and demonstrates that data efficiency can be improved with careful architectural choices rooted in the underlying physics.
M. A. Maia, K. Meyer, A. Van Gils et al.
· 1 citation