On maximum a posteriori estimation with Plug \& Play priors and stochastic gradient descent

DOI10.1007/S10851-022-01134-7arXiv2201.06133OpenAlexW4317379396MaRDI QIDQ6155451FDOQ6155451

Authors: Rémi Laumont, Valentin De Bortoli, Andrés Almansa, Julie Delon, Alain Durmus, Marcelo Pereyra 

Publication date: 12 June 2023

Published in: Journal of Mathematical Imaging and Vision (Search for Journal in Brave)

Abstract: Bayesian methods to solve imaging inverse problems usually combine an explicit data likelihood function with a prior distribution that explicitly models expected properties of the solution. Many kinds of priors have been explored in the literature, from simple ones expressing local properties to more involved ones exploiting image redundancy at a non-local scale. In a departure from explicit modelling, several recent works have proposed and studied the use of implicit priors defined by an image denoising algorithm. This approach, commonly known as Plug & Play (PnP) regularisation, can deliver remarkably accurate results, particularly when combined with state-of-the-art denoisers based on convolutional neural networks. However, the theoretical analysis of PnP Bayesian models and algorithms is difficult and works on the topic often rely on unrealistic assumptions on the properties of the image denoiser. This papers studies maximum-a-posteriori (MAP) estimation for Bayesian models with PnP priors. We first consider questions related to existence, stability and well-posedness, and then present a convergence proof for MAP computation by PnP stochastic gradient descent (PnP-SGD) under realistic assumptions on the denoiser used. We report a range of imaging experiments demonstrating PnP-SGD as well as comparisons with other PnP schemes.

Full work available at URL: https://arxiv.org/abs/2201.06133

Recommendations

• Approximate maximum a posteriori with Gaussian process priors
• scientific article; zbMATH DE number 6860839
• Learning maximum a-posteriori perturbation models
• Generalized gradients: priors on minimization flows
• A posteriori minimax estimation with likelihood constraints
• Learning to optimize via posterior sampling
• A Bayesian conjugate gradient method (with discussion)
• A Probabilistic Approach to Gradient Estimates
• A recursive maximum a posteriori estimator
• scientific article; zbMATH DE number 7219341

zbMATH Keywords

denoisinginverse problemsdeblurringinpaintingstochastic gradient descentBayesian imagingPlug and Play

Mathematics Subject Classification ID

Bayesian inference (62F15) Numerical mathematical programming methods (65K05) Numerical optimization and variational techniques (65K10) Bayesian problems; characterization of Bayes procedures (62C10) Analysis of algorithms and problem complexity (68Q25) Nonconvex programming, global optimization (90C26) Computing methodologies for image processing (68U10) Computer science (68-XX) Information and communication theory, circuits (94-XX)

Cites Work

• Scalable Bayesian uncertainty quantification in imaging inverse problems via convex optimization
• Nonlinear total variation based noise removal algorithms
• The Bayesian Choice
• Distributed optimization and statistical learning via the alternating direction method of multipliers
• Convex analysis and monotone operator theory in Hilbert spaces
• Convergence of a stochastic approximation version of the EM algorithm
• The Little Engine that Could: Regularization by Denoising (RED)
• Symmetrizing smoothing filters
• Solving Inverse Problems With Piecewise Linear Estimators: From Gaussian Mixture Models to Structured Sparsity
• A Review of Image Denoising Algorithms, with a New One
• De-noising by soft-thresholding
• Proximal splitting methods in signal processing
• An algorithm for total variation minimization and applications
• A first-order primal-dual algorithm for convex problems with applications to imaging
• Tweedie’s Formula and Selection Bias
• Applications of a Kushner and Clark lemma to general classes of stochastic algorithms
• Title not available (Why is that?)
• Title not available (Why is that?)
• Convex optimization: algorithms and complexity
• General results on the convergence of stochastic algorithms
• A nonlocal Bayesian image denoising algorithm
• Do stochastic algorithms avoid traps?
• Lectures on convex optimization
• Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising
• A Convergent Image Fusion Algorithm Using Scene-Adapted Gaussian-Mixture-Based Denoising
• Optimization methods for large-scale machine learning
• Solving inverse problems using data-driven models
• Breaking the curse of dimensionality with convex neural networks
• Should Penalized Least Squares Regression be Interpreted as Maximum A Posteriori Estimation?
• Posterior expectation of the total variation model: properties and experiments
• Accelerating Proximal Markov Chain Monte Carlo by Using an Explicit Stabilized Method
• Asymptotic bias of stochastic gradient search
• Maximum-a-posteriori estimation with Bayesian confidence regions
• Plug-and-Play Unplugged: Optimization-Free Reconstruction Using Consensus Equilibrium
• Efficient Bayesian computation by proximal Markov chain Monte Carlo: when Langevin meets Moreau
• Maximum likelihood estimation of regularization parameters in high-dimensional inverse problems: an empirical Bayesian approach. I: Methodology and experiments
• Learning maximally monotone operators for image recovery
• DeepISP: Toward Learning an End-to-End Image Processing Pipeline
• Bayesian Imaging Using Plug & Play Priors: When Langevin Meets Tweedie

Cited In (1)

• Averaged deep denoisers for image regularization

This page was built for publication: On maximum a posteriori estimation with Plug \& Play priors and stochastic gradient descent

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6155451)