∂-NO
An architecture-agnostic derivative-enhanced neural operator that improves PDE solution accuracy by supplying learnable fractional-order derivative features to any neural operator backbone.
Timeline (last 90 days)
The study proposes ∂-NO (del-NO), an architecture-agnostic plug-and-play derivative-enhanced neural operator that can provide learnable fractional-order derivative features for any neural operator backbone.
Experiments on multiple benchmark problems and different architectures show that ∂-NO consistently improves prediction accuracy.
The learned order reflects a modulated representation of known physical laws under the constraints of noise and limited samples, indicating that the network learns not only a numerical fit but also captures physical structure.
FNO, DeepONet, or other neural operator architectures can all directly integrate the ∂-NO enhancement without redesigning the network structure.
All Facts (4)
Experiments on multiple benchmark problems and different architectures show that ∂-NO consistently improves prediction accuracy.
50%UnverifiedThe learned order reflects a modulated representation of known physical laws under the constraints of noise and limited samples, indicating that the network learns not only a numerical fit but also captures physical structure.
50%UnverifiedFNO, DeepONet, or other neural operator architectures can all directly integrate the ∂-NO enhancement without redesigning the network structure.
50%UnverifiedThe study proposes ∂-NO (del-NO), an architecture-agnostic plug-and-play derivative-enhanced neural operator that can provide learnable fractional-order derivative features for any neural operator backbone.
50%