Can Data-Driven Dynamics Reveal Hidden Physics? There Is A Need for Interpretable Neural Operators
Not on the map. Why?
Abstract (arXiv)
Recently, neural operators have emerged as powerful tools for learning mappings between function spaces, enabling data-driven simulations of complex dynamics. Despite their successes, a deeper understanding of their learning mechanisms remains underexplored. In this work, we classify neural operators into two types: (1) Spatial domain models that learn on grids and (2) Functional domain models that learn with function bases. We present several viewpoints based on this classification and focus on learning data-driven dynamics adhering to physical principles. Specifically, we provide a way to explain the prediction-making process of neural operators and show that neural operator can learn hidden physical patterns from data. However, this explanation method is limited to specific situations, highlighting the urgent need for generalizable explanation methods. Next, we show that a simple dual-space multi-scale model can achieve SOTA performance and we believe that dual-space multi-spatio-scale models hold significant potential to learn complex physics and require further investigation. Lastly, we discuss the critical need for principled frameworks to incorporate known physics into neural operators, enabling better generalization and uncovering more hidden physical phenomena.
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 27 of its 45 references that are in the map. First posted on arXiv 3 Oct 2025.
Builds on
27 of 45 references.
References · 27
- 2020Fourier Neural Operator for Parametric Partial Differential Equationscited by 2,806
- 2019DeepONet: Learning nonlinear operators for identifying differential equations based on the universal approximation theorem of operatorscited by 1,965
- 2016Group Equivariant Convolutional Networkscited by 1,154
- 2022FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operatorscited by 748
- 2017Understanding the Effective Receptive Field in Deep Convolutional Neural Networkscited by 549
- 2021Learning the solution operator of parametric partial differential equations with physics-informed DeepOnetscited by 537
- 2021Physics-Informed Neural Operator for Learning Partial Differential Equationscited by 516
- 2021Choose a Transformer: Fourier or Galerkincited by 333
- 2023GNOT: A General Neural Operator Transformer for Operator Learningcited by 321
- 2024Transolver: A Fast Transformer Solver for PDEs on General Geometriescited by 299
- 2020Self-Adaptive Physics-Informed Neural Networks using a Soft Attention Mechanismcited by 231
- 2022Towards Multi-spatiotemporal-scale Generalized PDE Modelingcited by 190
15 more
- 2024DPOT: Auto-Regressive Denoising Operator Transformer for Large-Scale PDE Pre-Trainingcited by 137
- 2020Incorporating Symmetry into Deep Dynamics Models for Improved Generalizationcited by 123
- 2023Scalable Transformer for PDE Surrogate Modelingcited by 123
- 2023Artificial Intelligence for Science in Quantum, Atomistic, and Continuum Systemscited by 118
- 2024Neural Operators with Localized Integral and Differential Kernelscited by 59
- 2023Improved Operator Learning by Orthogonal Attentioncited by 58
- 2023Representation Equivalent Neural Operators: a Framework for Alias-free Operator Learningcited by 57
- 2023Group Equivariant Fourier Neural Operators for Partial Differential Equationscited by 42
- 2024Equivariant Frames and the Impossibility of Continuous Canonicalizationcited by 41
- 2022INO: Invariant Neural Operators for Learning Complex Physical Systems with Momentum Conservationcited by 28
- 2022Transform Once: Efficient Operator Learning in Frequency Domaincited by 20
- 2024Position: Optimization in SciML Should Employ the Function Space Geometrycited by 20
- 2024Discretization Error of Fourier Neural Operatorscited by 20
- 2024Nonlocal Attention Operator: Materializing Hidden Knowledge Towards Interpretable Physics Discoverycited by 18
- 2024Positional Knowledge is All You Need: Position-induced Transformer (PiT) for Operator Learningcited by 17
Built on by
Citing papers per year: 2025: 1.
1 paper.