Deep convolutional Ritz method: parametric PDE surrogates without labeled data
From MaRDI portal
Publication:6132294
Abstract: Parametric surrogate models for partial differential equations (PDEs) are a necessary component for many applications in the computational sciences, and convolutional neural networks (CNNs) have proved as an excellent tool to generate these surrogates when parametric fields are present. CNNs are commonly trained on labeled data based on one-to-one sets of parameter-input and PDE-output fields. Recently, residual-based convolutional physics-informed neural network (CPINN) solvers for parametric PDEs have been proposed to build surrogates without the need for labeled data. These allow for the generation of surrogates without an expensive offline-phase. In this work, we present an alternative formulation termed Deep Convolutional Ritz Method (DCRM) as a parametric PDE solver. The approach is based on the minimization of energy functionals, which lowers the order of the differential operators compared to residual-based methods. Based on studies involving the Poisson equation with a spatially parameterized source term and boundary conditions, we found that CNNs trained on labeled data outperform CPINNs in convergence speed and generalization ability. Surrogates generated from DCRM, however, converge significantly faster than their CPINN counterparts and prove to generalize faster and better than surrogates obtained from both CNNs trained on labeled data and CPINNs. This hints that DCRM could make PDE solution surrogates trained without labeled data possible.
Cites work
- scientific article; zbMATH DE number 6378127 (Why is no real title available?)
- scientific article; zbMATH DE number 4141383 (Why is no real title available?)
- scientific article; zbMATH DE number 5297089 (Why is no real title available?)
- scientific article; zbMATH DE number 3464364 (Why is no real title available?)
- scientific article; zbMATH DE number 1181255 (Why is no real title available?)
- scientific article; zbMATH DE number 2143187 (Why is no real title available?)
- scientific article; zbMATH DE number 5180707 (Why is no real title available?)
- A Gaussian process regression approach within a data-driven POD framework for engineering problems in fluid dynamics
- A global version of the inverse problem of the calculus of variations
- A heteroencoder architecture for prediction of failure locations in porous metals using variational inference
- A physics-informed neural network technique based on a modified loss function for computational 2D and 3D solid mechanics
- A reduced multiscale model for nonlinear structural topology optimization
- A theoretical analysis of deep neural networks and parametric PDEs
- An energy approach to the solution of partial differential equations in computational mechanics via machine learning: concepts, implementation and applications
- Bayesian deep convolutional encoder-decoder networks for surrogate modeling and uncertainty quantification
- Calibrated reduced-order POD-Galerkin system for fluid flow modelling
- Convergence rate analysis for deep Ritz method
- Deep Nitsche Method: Deep Ritz Method with Essential Boundary Conditions
- Deep UQ: learning deep neural network surrogate models for high dimensional uncertainty quantification
- Improving the efficiency of large scale topology optimization through on-the-fly reduced order model construction
- Introduction to linear regression analysis
- Large-scale PDE-constrained optimization
- Local approximate Gaussian process regression for data-driven constitutive models: development and comparison with neural networks
- Model reduction and neural networks for parametric PDEs
- Model-data-driven constitutive responses: application to a multiscale computational framework
- Nonlinear finite element methods
- On physics-informed data-driven isotropic and anisotropic constitutive models through probabilistic machine learning and space-filling sampling
- On the Existence of Global Variational Principles
- POLYNOMIAL-CHAOS-BASED KRIGING
- PhyCRNet: physics-informed convolutional-recurrent network for solving spatiotemporal PDEs
- PhyGeoNet: physics-informed geometry-adaptive convolutional neural networks for solving parameterized steady-state PDEs on irregular domain
- 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
- Progressive construction of a parametric reduced-order model for PDE-constrained optimization
- Reduced basis methods for uncertainty quantification
- Reduced order modeling for nonlinear structural analysis using Gaussian process regression
- Reducing the Dimensionality of Data with Neural Networks
- Robust topology optimization with low rank approximation using artificial neural networks
- Solution of the Inverse Problem of the Calculus of Variations
- Solving parametric PDE problems with artificial neural networks
- The Hitchhiker's guide to the virtual element method
- The Inverse Problem of the Calculus of Variations
- The deep Ritz method: a deep learning-based numerical algorithm for solving variational problems
- The finite volume method in computational fluid dynamics. An advanced introduction with OpenFOAM and Matlab
- The mixed deep energy method for resolving concentration features in finite strain hyperelasticity
- The neural particle method - an updated Lagrangian physics informed neural network for computational fluid dynamics
- When and why PINNs fail to train: a neural tangent kernel perspective
This page was built for publication: Deep convolutional Ritz method: parametric PDE surrogates without labeled data
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6132294)
