AI summaryⓘ
The authors address the problem where neural operator models trained on simulated data perform poorly on real experimental data. They propose PhysGuard, a method that protects the important physics-related parts of the model during fine-tuning so it adapts better to real data without losing key physics knowledge. PhysGuard uses a mathematical tool called the Fisher Information Matrix to figure out which parts of the model are critical, then limits updates to less important parts. Their tests show that PhysGuard helps reduce errors, especially when the simulated and real data are very different. This approach improves the model's ability to generalize to real-world physics problems while keeping important learned structures intact.
Neural operatorsSim-to-real gapFine-tuningFisher Information MatrixPhysics-preserving adaptationDomain shiftLow-frequency structuresGram matrixSpectral probe
Authors
Changjian Zhou, Junfeng Fang, Negin Yousefpour, Peng Wu, Bin Yan, Guillermo A Narsilio
Abstract
Neural operator models trained on simulation data often lose accuracy when applied to experimental measurements due to the sim-to-real gap. Standard fine-tuning with limited real data can reduce this gap, but it may also damage the core physics-relevant representations learned during pretraining. Although knowledge-preserving adaptation has been widely investigated in vision or language tasks, it remains unclear whether these methods are suitable for neural operators whose architectures and protected knowledge are fundamentally different. Neural operators need to preserve core-scale physical structures rather than semantic or visual features. We propose PhysGuard, a physics-preserving framework for accurate sim-to-real adaptation of neural operators. Specifically, PhysGuard uses the empirical Fisher Information Matrix computed on simulation data to identify physics-critical parameter directions, then restricts fine-tuning updates to directions that do not interfere with them. A layer-wise Gram-matrix formulation makes this efficient for models with millions of parameters, while an adaptive threshold automatically determines the protected subspace size. A spectral probe experiment shows that the dominant Fisher directions are strongly associated with low-frequency output structures. Experiments on benchmark across four neural operator architectures and different physical systems show that PhysGuard performs strongly on most evaluation metrics compared to baselines. The benefits are most evident under severe domain shift, where it reduces low-frequency error by up to 32\% compared to standard fine-tuning while maintaining adaptability. Our code is available at https://github.com/ZhouChaunge/PhysGuard.