MgNet: A Unified Framework of Multigrid and Convolutional Neural Network

Abstract

We develop a unified model, known as MgNet, that simultaneously recovers some convolutional neural networks (CNN) for image classification and multigrid (MG) methods for solving discretized partial differential equations (PDEs). This model is based on close connections that we have observed and uncovered between the CNN and MG methodologies. For example, pooling operation and feature extraction in CNN correspond directly to restriction operation and iterative smoothers in MG, respectively. As the solution space is often the dual of the data space in PDEs, the analogous concept of feature space and data space (which are dual to each other) is introduced in CNN. With such connections and new concept in the unified model, the function of various convolution operations and pooling used in CNN can be better understood. As a result, modified CNN models (with fewer weights and hyper parameters) are developed that exhibit competitive and sometimes better performance in comparison with existing CNN models when applied to both CIFAR-10 and CIFAR-100 data sets.

30 pages

References in corpus (10)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification
• Learning Deconvolution Network for Semantic Segmentation
• PDE-Net 2.0: Learning PDEs from Data with A Numeric-Symbolic Hybrid Deep Network
• The Reversible Residual Network: Backpropagation Without Storing Activations
• Learning Neural PDE Solvers with Convergence Guarantees
• Dual Path Networks
• Approximation Rates for Neural Networks with General Activation Functions
• A Flow Model of Neural Networks

Cited by in corpus (30)

• PhyGeoNet: Physics-Informed Geometry-Adaptive Convolutional Neural Networks for Solving Parameterized Steady-State PDEs on Irregular Domain
• Local Extreme Learning Machines and Domain Decomposition for Solving Linear and Nonlinear Partial Differential Equations
• A Review on Deep Learning in Medical Image Reconstruction
• Multipole Graph Neural Operator for Parametric Partial Differential Equations
• PFNN: A Penalty-Free Neural Network Method for Solving a Class of Second-Order Boundary-Value Problems on Complex Geometries
• Probabilistic Deep Learning for Real-Time Large Deformation Simulations
• On Computing the Hyperparameter of Extreme Learning Machines: Algorithm and Application to Computational PDEs, and Comparison with Classical and High-Order Finite Elements
• Choose a Transformer: Fourier or Galerkin
• Physics-Informed Supervised Residual Learning for Electromagnetic Modeling
• A Modified Batch Intrinsic Plasticity Method for Pre-training the Random Coefficients of Extreme Learning Machines
• A Helmholtz equation solver using unsupervised learning: Application to transcranial ultrasound
• PocketNet: A Smaller Neural Network for Medical Image Analysis
• Numerical Approximation of Partial Differential Equations by a Variable Projection Method with Artificial Neural Networks
• Approximation Properties of Deep ReLU CNNs
• Variational operator learning: A unified paradigm marrying training neural operators and solving partial differential equations
• DNN-MG: A Hybrid Neural Network/Finite Element Method with Applications to 3D Simulations of the Navier-Stokes Equations
• An Interpretable MRI Reconstruction Network with Two-grid-cycle Correction and Geometric Prior Distillation
• Enforcing Deterministic Constraints on Generative Adversarial Networks for Emulating Physical Systems
• Multiresolution Convolutional Autoencoders
• Cell-average based neural network method for hyperbolic and parabolic partial differential equations
• Bayesian Sparse learning with preconditioned stochastic gradient MCMC and its applications
• Meta-MgNet: Meta Multigrid Networks for Solving Parameterized Partial Differential Equations
• Neural-network preconditioners for solving the Dirac equation in lattice gauge theory
• FAS-UNet: A Novel FAS-driven Unet to Learn Variational Image Segmentation
• Learning Green's Functions of Linear Reaction-Diffusion Equations with Application to Fast Numerical Solver
• Learning optimal multigrid smoothers via neural networks
• Accuracy and Architecture Studies of Residual Neural Network solving Ordinary Differential Equations
• Constrained Linear Data-feature Mapping for Image Classification
• A machine learning method correlating pulse pressure wave data with pregnancy
• Estimating Adsorption Isotherm Parameters in Chromatography via a Virtual Injection Promoting Feed-forward Neural Network