PDE Identification Using Noise Adaptive Differentiation in Strong Form (S-IDENT)
arXiv:2606.31776
2026
Architecture
1 ideas extracted · analyzed Aug 29, 2026
What the math gives to ML
The transferable contribution is a principled noisy-derivative front end rather than PDE identification itself. Savitzky-Golay polynomial differentiation provides explicit convolution filters that preserve low-order polynomial structure, while Stein's Unbiased Risk Estimate selects the local window by balancing residual error and noise amplification. This can replace fixed finite differences or learned derivative filters in temporal models, neural operators, and state-space networks. The main falsifiable benefit is more accurate and stable derivative information under observation noise at negligible inference cost.
Ideas from this paper
Unverified
2026
Prepend an adaptive Savitzky-Golay derivative bank to a temporal neural network. For each input channel and derivative order, select the local window by minimizing Stein's unbiased risk estimate, then concatenate the raw signal with the estimated derivatives. This supplies denoised velocity and acceleration features without requiring clean derivative targets or forcing the backbone to learn unstable finite-difference filters.
Useful5/10
Difficulty3/10
Novelty6/10