NETT: solving inverse problems with deep neural networks

DOI10.1088/1361-6420/AB6D57zbMATH Open1456.65038arXiv1803.00092OpenAlexW3002289736MaRDI QIDQ5113628FDOQ5113628

Authors: Housen Li, Johannes Schwab, Stephan Antholzer, Markus Haltmeier 

Publication date: 15 June 2020

Published in: Inverse Problems (Search for Journal in Brave)

Abstract: Recovering a function or high-dimensional parameter vector from indirect measurements is a central task in various scientific areas. Several methods for solving such inverse problems are well developed and well understood. Recently, novel algorithms using deep learning and neural networks for inverse problems appeared. While still in their infancy, these techniques show astonishing performance for applications like low-dose CT or various sparse data problems. However, there are few theoretical results for deep learning in inverse problems. In this paper, we establish a complete convergence analysis for the proposed NETT (Network Tikhonov) approach to inverse problems. NETT considers data consistent solutions having small value of a regularizer defined by a trained neural network. We derive well-posedness results and quantitative error estimates, and propose a possible strategy for training the regularizer. Our theoretical results and framework are different from any previous work using neural networks for solving inverse problems. A possible data driven regularizer is proposed. Numerical results are presented for a tomographic sparse data problem, which demonstrate good performance of NETT even for unknowns of different type from the training data. To derive the convergence and convergence rates results we introduce a new framework based on the absolute Bregman distance generalizing the standard Bregman distance from the convex to the non-convex case.

Full work available at URL: https://arxiv.org/abs/1803.00092

Recommendations

• Big in Japan: regularizing networks for solving inverse problems
• Solving ill-posed inverse problems using iterative deep neural networks
• Deep neural networks can stably solve high-dimensional, noisy, non-linear inverse problems
• Deep learning for trivial inverse problems
• Deep learning for inverse problems with unknown operator

zbMATH Keywords

deep learninginverse problemsconvergence analysisimage reconstructionconvolutional neural networksabsolute Bregman distance

Mathematics Subject Classification ID

Nonparametric estimation (62G05) Artificial neural networks and deep learning (68T07) Image processing (compression, reconstruction, etc.) in information and communication theory (94A08) Numerical solution to inverse problems in abstract spaces (65J22)

Cites Work

• $rm K$-SVD: An Algorithm for Designing Overcomplete Dictionaries for Sparse Representation
• A wavelet tour of signal processing. The sparse way.
• Orthonormal bases of compactly supported wavelets
• Deep learning
• The Little Engine that Could: Regularization by Denoising (RED)
• Proximal splitting methods in signal processing
• Totally convex functions for fixed points computation and infinite dimensional optimization
• Mathematical methods in image reconstruction
• Necessary and sufficient conditions for linear convergence of \(\ell^1\)-regularization
• Mathematics of thermoacoustic tomography
• An iterative thresholding algorithm for linear inverse problems with a sparsity constraint
• Convergence rates of convex variational regularization
• Title not available (Why is that?)
• Inversion of Spherical Means and the Wave Equation in Even Dimensions
• Variational methods in imaging
• New tight frames of curvelets and optimal representations of objects with piecewise C² singularities
• Title not available (Why is that?)
• The residual method for regularizing ill-posed problems
• A Tikhonov-based projection iteration for nonlinear ill-posed problems with sparsity con\-straints
• Sparse regularization with l ^q penalty term
• Dictionary Identification—Sparse Matrix-Factorization via $\ell_1$-Minimization
• Convergence rates and source conditions for Tikhonov regularization with sparsity constraints
• Generalized Bregman distances and convergence rates for non-convex regularization methods
• Temporal back-projection algorithms for photoacoustic tomography with integrating line detectors
• Solving ill-posed inverse problems using iterative deep neural networks
• Deep Convolutional Neural Network for Inverse Problems in Imaging
• Deep learning for photoacoustic tomography from sparse data
• Photoacoustic and thermoacoustic tomography: image formation principles
• Sampling Conditions for the Circular Radon Transform

Cited In (87)

• Adversarially learned iterative reconstruction for imaging inverse problems
• Big in Japan: regularizing networks for solving inverse problems
• Networks for nonlinear diffusion problems in imaging
• Learnable descent algorithm for nonsmooth nonconvex image reconstruction
• Deep learning methods for partial differential equations and related parameter identification problems
• Learning Regularization Parameter-Maps for Variational Image Reconstruction Using Deep Neural Networks and Algorithm Unrolling
• The numerics of phase retrieval
• Strong Solutions for PDE-Based Tomography by Unsupervised Learning
• A direct sampling-based deep learning approach for inverse medium scattering problems
• DRIP: deep regularizers for inverse problems
• Deep learning architectures for nonlinear operator functions and nonlinear inverse problems
• Wide-band butterfly network: stable and efficient inversion via multi-frequency neural networks
• Convolutional Neural Networks in Phase Space and Inverse Problems
• Shared Prior Learning of Energy-Based Models for Image Reconstruction
• Deep-learning based discovery of partial differential equations in integral form from sparse and noisy data
• Deep convolutional framelets: a general deep learning framework for inverse problems
• A mathematical framework for deep learning in elastic source imaging
• Analytic continuation of noisy data using Adams Bashforth residual neural network
• Deep learning for inverse problems. Abstracts from the workshop held March 7--13, 2021 (hybrid meeting)
• Deep neural networks for inverse problems with pseudodifferential operators: an application to limited-angle tomography
• Gauss-Newton method for solving linear inverse problems with neural network coders
• Wasserstein-based projections with applications to inverse problems
• Deep Learning--Based Dictionary Learning and Tomographic Image Reconstruction
• Divide-and-conquer DNN approach for the inverse point source problem using a few single frequency measurements
• Convergence analysis of critical point regularization with non-convex regularizers
• Deep neural networks can stably solve high-dimensional, noisy, non-linear inverse problems
• Uniformly convex neural networks and non-stationary iterated network Tikhonov (iNETT) method
• NESTANets: stable, accurate and efficient neural networks for analysis-sparse inverse problems
• Deep null space learning for inverse problems: convergence analysis and rates
• Deep learning for inverse problems with unknown operator
• Regularization theory of the analytic deep prior approach
• Adaptive Tikhonov strategies for stochastic ensemble Kalman inversion
• Deep synthesis network for regularizing inverse problems
• Learning regularization parameters of inverse problems via deep neural networks
• Relaxation approach for learning neural network regularizers for a class of identification problems
• Hybrid projection methods for large-scale inverse problems with mixed Gaussian priors
• Parseval proximal neural networks
• Compressive sensing and neural networks from a statistical learning perspective
• Solving ill-posed inverse problems using iterative deep neural networks
• Computed tomography reconstruction using deep image prior and learned reconstruction methods
• NETT
• Invertible residual networks in the context of regularization theory for linear inverse problems
• Data driven regularization by projection
• Bayesian Imaging with Data-Driven Priors Encoded by Neural Networks
• Solving an inverse source problem by deep neural network method with convergence and error analysis
• Learnable Douglas-Rachford iteration and its applications in DOT imaging
• Data-consistent neural networks for solving nonlinear inverse problems
• A Generative Variational Model for Inverse Problems in Imaging
• Estimating adsorption isotherm parameters in chromatography via a virtual injection promoting double feed-forward neural network
• Numerical solution of inverse problems by weak adversarial networks
• On learned operator correction in inverse problems
• Solving inverse problems using data-driven models
• Linearized inverse Schrödinger potential problem with partial data and its deep neural network inversion
• Limitations of neural network training due to numerical instability of backpropagation
• Deep learning for trivial inverse problems
• ADMM based Fourier phase retrieval with untrained generative prior
• Deep unrolling networks with recurrent momentum acceleration for nonlinear inverse problems
• Uncertainty quantification for goal-oriented inverse problems via variational encoder-decoder networks
• The Calderón's problem via DeepONets
• Convergent data-driven regularizations for CT reconstruction
• Alpert multi-wavelets for functional inverse problems: direct optimization and deep learning
• On obtaining initial approximation for the full wave inversion problem using convolutional neural network
• An accelerated inexact Newton regularization scheme with a learned feature-selection rule for non-linear inverse problems
• Regularising inverse problems with generative machine learning models
• TNet: A Model-Constrained Tikhonov Network Approach for Inverse Problems
• Rethinking the ill-posedness of the spectral function reconstruction -- why is it fundamentally hard and how artificial neural networks can help
• Training adaptive reconstruction networks for blind inverse problems
• Bregman methods for large-scale optimization with applications in imaging
• Bilevel optimization methods in imaging
• Deep learning methods for limited data problems in X-ray tomography
• Learned regularizers for inverse problems
• Regularization of inverse problems by neural networks
• Application of machine learning regression models to inverse eigenvalue problems
• Iteratively Refined Image Reconstruction with Learned Attentive Regularizers
• Bayesian view on the training of invertible residual networks for solving linear inverse problems
• Quadratic Neural Networks for Solving Inverse Problems
• Numerical methods for identifying the diffusion coefficient in a nonlinear elliptic equation
• Solving inverse problems with deep learning
• Oracle-net for nonlinear compressed sensing in electrical impedance tomography reconstruction problems
• Solving Bayesian inverse problems using neural networks
• Proximal neural networks and stochastic normalizing flows for inverse problems
• Normalizing flow regularization for photoacoustic tomography
• Data-Driven Morozov Regularization of Inverse Problems
• Learning spiking neuronal networks with artificial neural networks: neural oscillations
• Neural-network-based regularization methods for inverse problems in imaging
• Robustness and exploration of variational and machine learning approaches to inverse problems: an overview
• Learning weakly convex regularizers for convergent image-reconstruction algorithms

Uses Software

• UNLocBoX
• U-Net

This page was built for publication: NETT: solving inverse problems with deep neural networks

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5113628)