Abstract:We consider non-linear Bayesian inverse problems of determining the parameter $f$. For the posterior distribution with a class of Gaussian process priors, we study the statistical performance of variational Bayesian inference to the posterior with variational sets consisting of Gaussian measures or a mean-field family. We propose certain conditions on the forward map $\mathcal{G}$, the variational set $\mathcal{Q}$ and the prior such that, as the number $N$ of measurements increases, the resulting variational posterior distributions contract to the ground truth $f_0$ generating the data, and derive a convergence rate with polynomial order or logarithmic order. As specific examples, we consider a collection of non-linear inverse problems, including the Darcy flow problem, the inverse potential problem for a subdiffusion equation, and the inverse medium scattering problem. Besides, we show that our convergence rates are minimax optimal for these inverse problems.

What problem does this paper attempt to address?

This paper attempts to solve the convergence problem of variational posterior distributions in nonlinear Bayesian inverse problems. Specifically, the author studies how the variational posterior distribution obtained by the variational Bayes inference method shrinks to the true parameters as the amount of measurement data increases, given a Gaussian process prior. The paper proposes some conditions regarding the forward mapping $G$, the variational set $Q$, and the prior, which ensure that the variational posterior distribution converges to the true parameters at a polynomial or logarithmic rate. In addition, the paper also explores the applications of these results in specific nonlinear inverse problems, such as Darcy flow problems, inverse problems of sub - diffusion equations, and inverse problems of medium scattering, and proves that these convergence rates are optimal (up to a logarithmic factor). ### Main contributions of the paper 1. **Theoretical framework**: A theoretical framework for the convergence of variational posterior distributions in nonlinear inverse problems is established, providing general conditions regarding the forward mapping $G$, the prior, and the variational set $Q$. 2. **Convergence rate**: The convergence rate of the variational posterior distribution is derived, and it is proved that these convergence rates are optimal in some cases. 3. **Specific applications**: The theoretical results are applied to specific nonlinear inverse problems, such as Darcy flow problems and inverse problems of sub - diffusion equations, to verify the effectiveness of the theory. ### Key formulas - **Optimization problem of variational posterior distribution**: \[ \hat{Q}=\arg\min_{Q\in\mathcal{Q}}\mathcal{D}_{\text{KL}}(Q\parallel\Pi(\cdot\mid\mathcal{Y}_n)) \] where $\mathcal{D}_{\text{KL}}$ represents the Kullback - Leibler divergence, and $\Pi(\cdot\mid\mathcal{Y}_n)$ is the posterior distribution based on the data $\mathcal{Y}_n$. - **Convergence rate**: \[ \epsilon_n^2+\frac{1}{n}\inf_{Q\in\mathcal{Q}}\mathcal{D}_{\text{KL}}(Q\parallel\Pi_n(\cdot\mid\mathcal{Y}_n)) \] where $\epsilon_n$ represents the convergence rate of the posterior distribution, and $\frac{1}{n}\inf_{Q\in\mathcal{Q}}\mathcal{D}_{\text{KL}}(Q\parallel\Pi_n(\cdot\mid\mathcal{Y}_n))$ represents the variational approximation error. ### Conclusion Through rigorous mathematical analysis, the paper proves that under appropriate conditions, the variational Bayes inference method can effectively estimate the parameters in nonlinear inverse problems, and the convergence speed of the variational posterior distribution is comparable to that of the traditional posterior distribution. This provides a theoretical support for efficient computation in practical applications.

Consistency of variational Bayesian inference for non-linear inverse problems of partial differential equations

Bayesian Inference and Deep Learning for Inverse Problems

Variational Gaussian Processes For Linear Inverse Problems

Consistency of Bayesian inference with Gaussian process priors for a parabolic inverse problem

Consistency of Bayesian inference with Gaussian process priors in an elliptic inverse problem

On the Convergence of Extended Variational Inference for Non-Gaussian Statistical Models.

VARIATIONAL BAYES' METHOD FOR FUNCTIONS WITH APPLICATIONS TO SOME INVERSE PROBLEMS

Variational Inverting Network for Statistical Inverse Problems of Partial Differential Equations

Efficient Variational Bayesian Model Updating by Bayesian Active Learning

A posterior contraction for Bayesian inverse problems in Banach spaces

On the Frequentist Coverage of Bayes Posteriors in Nonlinear Inverse Problems

Learning to solve Bayesian inverse problems: An amortized variational inference approach using Gaussian and Flow guides

Statistical guarantees for Bayesian uncertainty quantification in non-linear inverse problems with Gaussian process priors

Variational Bayes algorithm and posterior consistency of Ising model parameter estimation

A variational inference framework for inverse problems

Non-centered Parametric Variational Bayes' Approach for Hierarchical Inverse Problems of Partial Differential Equations.

Linear methods for non-linear inverse problems

A Bayesian approach with Gaussian priors to the inverse problem of source identification in elliptic PDEs

Variational Bayesian Inversion for the Reaction Coefficient in Space-Time Nonlocal Diffusion Equations

Solving linear Bayesian inverse problems using a fractional total variation-Gaussian (FTG) prior and transport map