物理约束 ML论文趋势 2023–2026

主要结论

• 物理约束 ML论文从 2023 年的 50 篇增长到 2026 年的 199 篇，三年间扩大到约 4.0 倍。
• NeurIPS 是发物理约束 ML论文最多的会议，2023–2026 年共 167 篇，占四个会议总量的 35%。
• 交叉方向里涨得最快的是多模态大模型：同时属于两个主题的论文从 2023 年的 0 篇增至 2026 年的 11 篇。
• 在「气候与物理科学」的 5 个主题里，它的 2026 年论文数排第 1；同组增长最快的天气预报是 6.1 倍，物理约束 ML是 4.0 倍。

代表论文

按年份和会议均匀挑选，方便快速看到每一届的样貌

• NeurIPS 2026Derivative-Informed Training of Neural Operators On-the-Fly via Sketched Tangent Consistency
• ICML 2026Bias-Spectrum Neural Processes for Parametric PDEs: Architecture Priors Meet PDE Constraints
• ICLR 2026CONSIGN: Conformal Segmentation Informed by Spatial Groupings via Decomposition
• CVPR 2026Domain Sensitive Federated Learning with Fisher-Informed Pruning
• NeurIPS 2025Activation-Informed Merging of Large Language Models
• ICML 2025Geometry-Informed Neural Networks
• ICLR 2025Beyond the convexity assumption: Realistic tabular data generation under quantifier-free real linear constraints
• CVPR 2025A Physics-Informed Blur Learning Framework for Imaging Systems
• NeurIPS 2024ChaosBench: A Multi-Channel, Physics-Based Benchmark for Subseasonal-to-Seasonal Climate Prediction
• ICML 2024A Space Group Symmetry Informed Network for O(3) Equivariant Crystal Tensor Prediction
• ICLR 2024AirPhyNet: Harnessing Physics-Guided Neural Networks for Air Quality Prediction
• CVPR 2024A Physics-informed Low-rank Deep Neural Network for Blind and Universal Lens Aberration Correction
• NeurIPS 2023ClimSim: A large multi-scale dataset for hybrid physics-ML climate emulation
• ICML 2023Gradient Descent Finds the Global Optima of Two-Layer Physics-Informed Neural Networks
• ICLR 2023Bi-level Physics-Informed Neural Networks for PDE Constrained Optimization using Broyden's Hypergradients
• CVPR 20233D Human Pose Estimation via Intuitive Physics
• NeurIPS 2026Neural Modal Decomposition: Architectural Priors from Observables
• ICML 2026Fine-to-Coarse Fairness-Informed Multi-View Clustering
• ICLR 2026From Neural Networks to Logical Theories: The Correspondence between Fibring Modal Logics and Fibring Neural Networks
• CVPR 2026LaDy Lagrangian-Dynamic Informed Network for Skeleton-based Action Segmentation via Spatial-Temporal

全部物理约束 ML论文

共 143 篇，第 2 / 3 页

51. ICLR 2026Rapid Training of Hamiltonian Graph Networks Using Random Features
52. ICLR 2026Representing local protein environments with machine learning force fields
53. ICLR 2026Sample-efficient evidence estimation of score based priors for model selection
54. ICLR 2026Sharing State Between Prompts and Programs
55. ICLR 2026SONATA: Synergistic Coreset Informed Adaptive Temporal Tensor Factorization
56. ICLR 2026STDDN: A Physics-Guided Deep Learning Framework for Crowd Simulation
57. ICLR 2026Structural Inference: Interpreting Small Language Models with Susceptibilities
58. ICLR 2026Towards Persistent Noise-Tolerant Active Learning of Regular Languages with Class Query
59. ICLR 2026Transformers are Inherently Succinct
60. ICLR 2026Two failure modes of deep transformers and how to avoid them: a unified theory of signal propagation at initialisation
61. ICLR 2026Uncertainty-Aware Diagnostics for Physics-Informed Machine Learning
62. ICLR 2026UrbanGraph: Physics-Informed Spatio-Temporal Dynamic Heterogeneous Graphs for Urban Microclimate Prediction
63. ICLR 2026Weight-Space Linear Recurrent Neural Networks
64. ICLR 2026WINA: Weight Informed Neuron Activation for Accelerating Large Language Model Inference
65. ICLR 2025ANaGRAM: A Natural Gradient Relative to Adapted Model for efficient PINNs learning
66. ICLR 2025Bayesian Optimization of Antibodies Informed by a Generative Model of Evolving Sequences
67. ICLR 2025Beyond the convexity assumption: Realistic tabular data generation under quantifier-free real linear constraints
68. ICLR 2025Circuit Transformer: A Transformer That Preserves Logical Equivalence
69. ICLR 2025ConFIG: Towards Conflict-free Training of Physics Informed Neural Networks
70. ICLR 2025Deep Learning Alternatives Of The Kolmogorov Superposition Theorem
71. ICLR 2025Differentiable Rule Induction from Raw Sequence Inputs
72. ICLR 2025Dream to Manipulate: Compositional World Models Empowering Robot Imitation Learning with Imagination
73. ICLR 2025Generating Physical Dynamics under Priors
74. ICLR 2025KLay: Accelerating Arithmetic Circuits for Neurosymbolic AI
75. ICLR 2025Learning a Neural Solver for Parametric PDEs to Enhance Physics-Informed Methods
76. ICLR 2025Learning Chaos In A Linear Way
77. ICLR 2025Learning system dynamics without forgetting
78. ICLR 2025Learning to Solve Differential Equation Constrained Optimization Problems
79. ICLR 2025Lie Algebra Canonicalization: Equivariant Neural Operators under arbitrary Lie Groups
80. ICLR 2025Logically Consistent Language Models via Neuro-Symbolic Integration
81. ICLR 2025MeshMask: Physics-Based Simulations with Masked Graph Neural Networks
82. ICLR 2025Metamizer: A Versatile Neural Optimizer for Fast and Accurate Physics Simulations
83. ICLR 2025NeSyC: A Neuro-symbolic Continual Learner For Complex Embodied Tasks In Open Domains
84. ICLR 2025Neural Stochastic Differential Equations for Uncertainty-Aware Offline RL
85. ICLR 2025Neural Wave Equation for Irregularly Sampled Sequence Data
86. ICLR 2025Physics of Language Models: Part 2.2, How to Learn From Mistakes on Grade-School Math Problems
87. ICLR 2025Physics-aligned field reconstruction with diffusion bridge
88. ICLR 2025Physics-Informed Diffusion Models
89. ICLR 2025Physics-informed Temporal Difference Metric Learning for Robot Motion Planning
90. ICLR 2025PIED: Physics-Informed Experimental Design for Inverse Problems
91. ICLR 2025PIG: Physics-Informed Gaussians as Adaptive Parametric Mesh Representations
92. ICLR 2025PINP: Physics-Informed Neural Predictor with latent estimation of fluid flows
93. ICLR 2025PIORF: Physics-Informed Ollivier-Ricci Flow for Long–Range Interactions in Mesh Graph Neural Networks
94. ICLR 2025PRDP: Progressively Refined Differentiable Physics
95. ICLR 2025Predicting the Energy Landscape of Stochastic Dynamical System via Physics-informed Self-supervised Learning
96. ICLR 2025Rethinking the role of frames for SE(3)-invariant crystal structure modeling
97. ICLR 2025Robust Weight Initialization for Tanh Neural Networks with Fixed Point Analysis
98. ICLR 2025Second-Order Fine-Tuning without Pain for LLMs: A Hessian Informed Zeroth-Order Optimizer
99. ICLR 2025Solving Differential Equations with Constrained Learning
100. ICLR 2025Student-Informed Teacher Training

常见问题

2023–2026 年四大会议一共有多少篇物理约束 ML论文？

共 481 篇，按年份分别是 2023 年 50 篇、2024 年 85 篇、2025 年 147 篇、2026 年 199 篇。

NeurIPS 2026 有多少篇物理约束 ML论文？

NeurIPS 2026 有 54 篇物理约束 ML论文，上一届是 65 篇。

ICLR 2026 有多少篇物理约束 ML论文？

ICLR 2026 有 64 篇物理约束 ML论文，上一届是 38 篇。

ICML 2026 有多少篇物理约束 ML论文？

ICML 2026 有 51 篇物理约束 ML论文，上一届是 25 篇。

CVPR 2026 有多少篇物理约束 ML论文？

CVPR 2026 有 30 篇物理约束 ML论文，上一届是 19 篇。

哪个会议的物理约束 ML论文最多？

NeurIPS，2023–2026 年共 167 篇。