mlsanityData to 7 Oct 2026

Stochastic Dimension Implicit Functional Projections for Global Integral Conservation in High-Dimensional PINNs

arXiv 2603.29237 ↗First posted 31 Mar 2026cs.LG cs.NA math.NAcited by 0

Not on the map. Why?

Abstract (arXiv)

Enforcing prescribed global integral constraints in mesh-free neural PDE solvers is challenging in high-dimensional domains. Existing projection methods for spatial integrals are often tied to fixed grids or uniform quadrature, which can conflict with randomly sampled physics-informed neural networks (PINNs) and scale poorly with dimension. High-order differential operators also increase reverse-mode automatic differentiation memory costs. We propose Stochastic Dimension Implicit Functional Projection (SDIFP), a quadrature-level framework for enforcing prescribed first and second spatial moments. SDIFP replaces tensor-product nodal projection by a global affine correction of the neural-network output, with two scalar coefficients determined from a weighted quadrature rule. Under positive target variance and nonzero empirical raw variance, this correction is the nearest-point projection, in the weighted quadrature norm, onto the empirical two-moment constraint set. Thus, the prescribed moments are exact for the selected quadrature rule, while continuum errors are quadrature errors of the corrected field. For decomposable high-dimensional linear operators, SDIFP combines affine moment correction with stochastic operator-subset sampling. With independent residual and derivative sampling and conditionally unbiased coefficient-gradient estimation, the resulting estimator is unbiased for the specified quadrature-based residual objective; the shared-subset fast mode is biased in general. SDIFP avoids tensor-product quadrature for moment enforcement, separates forward quadrature evaluation from the reverse-mode graph, and retains pointwise inference efficiency once the affine coefficients are fixed or precomputed.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 32 of its 52 references that are in the map. First posted on arXiv 31 Mar 2026.

Builds on

32 of 52 references.

References · 32

  1. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,160
  2. 2019Hamiltonian Neural Networkscited by 702
  3. 2019Differentiable Convex Optimization Layerscited by 458
  4. 2020Lagrangian Neural Networkscited by 388
  5. 2019Machine Learning for Fluid Mechanicscited by 380
  6. 2019PyHessian: Neural Networks Through the Lens of the Hessiancited by 291
  7. 2021DC3: A learning method for optimization with hard constraintscited by 181
  8. 2024Challenges in Training PINNs: A Loss Landscape Perspectivecited by 109
  9. 2019Physics-informed Autoencoders for Lyapunov-stable Fluid Flow Predictioncited by 67
  10. 2022Neural Conservation Laws: A Divergence-Free Perspectivecited by 67
  11. 2022Self-Supervised Primal-Dual Learning for Constrained Optimizationcited by 60
  12. 2020Efficient Learning of Generative Models via Finite-Difference Score Matchingcited by 49
20 more
  1. 2017Imposing Hard Constraints on Deep Networks: Promises and Limitationscited by 42
  2. 2021Physics and Equality Constrained Artificial Neural Networks: Application to Forward and Inverse Problems with Multi-fidelity Data Fusioncited by 38
  3. 2022Enhanced Physics-Informed Neural Networks with Augmented Lagrangian Relaxation Method (AL-PINNs)cited by 35
  4. 2021Universal Approximation Under Constraints is Possible with Transformerscited by 32
  5. 2023RAYEN: Imposition of Hard Convex Constraints on Neural Networkscited by 30
  6. 2024Physics-Informed Neural Networks with Hard Linear Equality Constraintscited by 30
  7. 2024HardNet: Hard-Constrained Neural Networks with Universal Approximation Guaranteescited by 29
  8. 2022Learning Physics-Informed Neural Networks without Stacked Back-propagationcited by 23
  9. 2023Hutchinson Trace Estimation for High-Dimensional and High-Order Physics-Informed Neural Networkscited by 23
  10. 2023Harnessing the Power of Neural Operators with Automatically Encoded Conservation Lawscited by 22
  11. 2024Score-Based Physics-Informed Neural Networks for High-Dimensional Fokker-Planck Equationscited by 22
  12. 2024Stochastic Taylor Derivative Estimator: Efficient amortization for arbitrary differential operatorscited by 20
  13. 2023Bias-Variance Trade-off in Physics-Informed Neural Networks with Randomized Smoothing for High-Dimensional PDEscited by 19
  14. 2024LinSATNet: The Positive Linear Satisfiability Neural Networkscited by 16
  15. 2025ENFORCE: Nonlinear Constrained Learning with Adaptive-depth Neural Projectioncited by 15
  16. 2023Exactly conservative physics-informed neural networks and deep operator networks for dynamical systemscited by 13
  17. 2024GLinSAT: The General Linear Satisfiability Neural Network Layer By Accelerated Gradient Descentcited by 7
  18. 2024Operator Splitting for Convex Constrained Markov Decision Processescited by 3
  19. 2026Hard Constraint Projection in a Physics Informed Neural Networkcited by 2
  20. 2023Douglas--Rachford is the best projection methodcited by 1

Built on by

Not cited yet.