Abstract:The Bayesian statistical framework provides a systematic approach to enhance the regularization model by incorporating prior information about the desired solution. For the Bayesian linear inverse problems with Gaussian noise and Gaussian prior, we propose a new iterative regularization algorithm that belongs to subspace projection regularization (SPR) methods. By treating the forward model matrix as a linear operator between the two underlying finite dimensional Hilbert spaces with new introduced inner products, we first introduce an iterative process that can generate a series of valid solution subspaces. The SPR method then projects the original problem onto these solution subspaces to get a series of low dimensional linear least squares problems, where an efficient procedure is developed to update the solutions of them to approximate the desired solution of the original problem. With the new designed early stopping rules, this iterative algorithm can obtain a regularized solution with a satisfied accuracy. Several theoretical results about the algorithm are established to reveal the regularization properties of it. We use both small-scale and large-scale inverse problems to test the proposed algorithm and demonstrate its robustness and efficiency. The most computationally intensive operations in the proposed algorithm only involve matrix-vector products, making it highly efficient for large-scale problems.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the regularization problem in large - scale Bayesian linear inverse problems. Specifically, the paper proposes an iterative regularization algorithm based on the Subspace Projection Regularization (SPR) framework to overcome the ill - posedness in large - scale Bayesian linear inverse problems. These problems usually occur in fields such as medical imaging, computed tomography, geoscience, and data assimilation. At its core is the reconstruction of unknown parameters or functions from indirect measurements in the presence of noise. ### Main Problems 1. **Ill - posedness**: Inverse problems are usually ill - posed, which means that there may be multiple solutions that can fit the observed data well, or the solution is very sensitive to small perturbations in the observed data. This makes it difficult for direct solution methods to obtain accurate solutions. 2. **Selection of regularization parameters**: In the regularization process, how to select appropriate regularization parameters is a challenge. Existing methods such as the L - curve method, Generalized Cross - Validation (GCV), and Discrepancy Principle (DP) all have their own limitations and require multiple solutions to the regularization problem. 3. **Computational efficiency**: For large - scale problems, traditional regularization methods require matrix inversion and decomposition, and these operations are very computationally expensive in high - dimensional situations. ### Solutions The paper proposes a new iterative regularization algorithm, which is based on the Subspace Projection Regularization (SPR) framework and uses the generalized Golub - Kahan bidiagonalization (gen - GKB) to generate an effective solution subspace. The specific steps are as follows: 1. **Subspace generation**: Generate a series of N - 1 orthogonal bases through the generalized Golub - Kahan bidiagonalization process, and these basis vectors form the solution subspace. 2. **Low - dimensional problem solving**: Project the original problem onto these solution subspaces to obtain a series of low - dimensional linear least - squares problems. 3. **Iterative update**: Design an efficient iterative update method to gradually update the solution and the residual norm without directly solving the low - dimensional problems. 4. **Early - stopping rules**: Propose several early - stopping rules based on the Discrepancy Principle (DP), L - curve (LC), and Generalized Cross - Validation (GCV) to avoid including too much noise components. ### Theoretical Contributions - **Mathematical framework**: Consider the matrix \(A\) as a bounded linear operator between two Hilbert spaces, which are equipped with \(N^{-1}\) and \(M^{-1}\) inner products respectively. - **Theoretical results**: Establish several theoretical results related to the regularization properties of the algorithm, including the explicit expression of the solution subspace and its connection with the optimal solution subspace, and prove that the iterative solution has a filtered Generalized Singular Value Decomposition (GSVD) expansion form. ### Experimental Results The paper tests the proposed algorithm through small - scale and large - scale inverse problems and compares it with the existing generalized hybrid methods. The experimental results show that the new algorithm is more robust and efficient in handling large - scale Bayesian linear inverse problems. In conclusion, the paper effectively solves the regularization and computational efficiency problems in large - scale Bayesian linear inverse problems by proposing an iterative algorithm based on subspace projection regularization.

Subspace projection regularization for large-scale Bayesian linear inverse problems

Bayesian Inference and Deep Learning for Inverse Problems

Projected Newton method for large-scale Bayesian linear inverse problems

A preconditioned Krylov subspace method for linear inverse problems with general-form Tikhonov regularization

Augmented Flexible Krylov Subspace methods with applications to Bayesian inverse problems

Generalized sparsity-promoting solvers for Bayesian inverse problems: Versatile sparsifying transforms and unknown noise variances

Sparse bayesian inference with regularized gaussian distributions

Regularization of Linear Inverse Problems with Irregular Noise Using Embedding Operators

Generalized Tikhonov regularization method for large-scale linear inverse problems

Linear Convergence of Inexact Descent Method and Inexact Proximal Gradient Algorithms for Lower-Order Regularization Problems

A posterior contraction for Bayesian inverse problems in Banach spaces

Bayesian Inference for Adaptive Low Rank and Sparse Matrix Estimation

Learning sparsity-promoting regularizers for linear inverse problems

A Joint Bidiagonalization Based Algorithm for Large Scale Linear Discrete Ill-posed Problems in General-Form Regularization

Prior normalization for certified likelihood-informed subspace detection of Bayesian inverse problems

Randomized Matrix Approximation to Enhance Regularized Projection Schemes in Inverse Problems

Bayesian High-dimensional Linear Regression with Sparse Projection-posterior

Sparse Bayesian Learning with Automatic-Weighting Laplace Priors for Sparse Signal Recovery.

Learning a Gaussian Mixture for Sparsity Regularization in Inverse Problems