Coercing machine learning to output physically accurate results
From MaRDI portal
Publication:2223280
DOI10.1016/J.JCP.2019.109099zbMATH Open1453.68164arXiv1910.09671OpenAlexW2985010153MaRDI QIDQ2223280FDOQ2223280
Authors: Yanyan Li 
Publication date: 28 January 2021
Published in: Journal of Computational Physics (Search for Journal in Brave)
Abstract: Many machine/deep learning artificial neural networks are trained to simply be interpolation functions that map input variables to output values interpolated from the training data in a linear/nonlinear fashion. Even when the input/output pairs of the training data are physically accurate (e.g. the results of an experiment or numerical simulation), interpolated quantities can deviate quite far from being physically accurate. Although one could project the output of a network into a physically feasible region, such a postprocess is not captured by the energy function minimized when training the network; thus, the final projected result could incorrectly deviate quite far from the training data. We propose folding any such projection or postprocess directly into the network so that the final result is correctly compared to the training data by the energy function. Although we propose a general approach, we illustrate its efficacy on a specific convolutional neural network that takes in human pose parameters (joint rotations) and outputs a prediction of vertex positions representing a triangulated cloth mesh. While the original network outputs vertex positions with erroneously high stretching and compression energies, the new network trained with our physics prior remedies these issues producing highly improved results.
Full work available at URL: https://arxiv.org/abs/1910.09671
Recommendations
- Theoretical prerequisites for physically justified machine learning and its applications to fluid dynamics
- A physics-guided neural network framework for elastic plates: comparison of governing equations-based and energy-based approaches
- Projection-based model reduction: formulations for physics-based machine learning
- Learning generative neural networks with physics knowledge
- Finite electro-elasticity with physics-augmented neural networks
Artificial neural networks and deep learning (68T07) Computer graphics; computational geometry (digital and algorithmic aspects) (68U05) Physics (00A79) Numerical aspects of computer graphics, image analysis, and computational geometry (65D18)
Cites Work
- Adaptive subgradient methods for online learning and stochastic optimization
- DGM: a deep learning algorithm for solving partial differential equations
- Title not available (Why is that?)
- On the limited memory BFGS method for large scale optimization
- A Stochastic Approximation Method
- Deep learning
- The Convergence of a Class of Double-rank Minimization Algorithms 1. General Considerations
- Coordinate descent algorithms
- A simple and efficient finite element for plate bending
- Some methods of speeding up the convergence of iteration methods
- A paradigm for data-driven predictive modeling using field inversion and machine learning
- Machine learning strategies for systems with invariance properties
- Hidden physics models: machine learning of nonlinear partial differential equations
- Physics-constrained deep learning for high-dimensional surrogate modeling and uncertainty quantification without labeled data
- Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
- Machine learning of linear differential equations using Gaussian processes
- Deep UQ: learning deep neural network surrogate models for high dimensional uncertainty quantification
- Identification of physical processes via combined data-driven and data-assimilation methods
- Deep learning algorithm for data-driven simulation of noisy dynamical system
- Sharp interface approaches and deep learning techniques for multiphase flows
- Enforcing constraints for interpolation and extrapolation in generative adversarial networks
Cited In (8)
- Convex optimization with an interpolation-based projection and its application to deep learning
- Symplectic neural networks in Taylor series form for Hamiltonian systems
- Neural Networks with Local Converging Inputs (NNLCI) for Solving Conservation Laws, Part I: 1D Problems
- On obtaining sparse semantic solutions for inverse problems, control, and neural network training
- Enforcing exact physics in scientific machine learning: a data-driven exterior calculus on graphs
- SDYN-GANs: adversarial learning methods for multistep generative models for general order stochastic dynamics
- Solving inverse-PDE problems with physics-aware neural networks
- Addressing discontinuous root-finding for subsequent differentiability in machine learning, inverse problems, and control
Uses Software
This page was built for publication: Coercing machine learning to output physically accurate results
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2223280)
