18 July 2022 in SIAM Journal on Imaging Sciences
In this paper, we focus on developing a fast Kaczmarz-type method to solve inverse problems that can be written as systems of linear or nonlinear equations in Hilbert spaces. In order to capture the special feature of solutions, we incorporate nonsmooth convex functions into the averaged Kaczmarz iteration, leading to a new Kaczmarz-type method. In addition, aimed at further accelerating our proposed method, the choice of the step size is carefully discussed. Under the similar assumptions of the Kaczmarz-type method, we prove that our method is a convergent regularization method as long as it is terminated by an appropriate stopping rule. Finally, detailed numerical studies are presented for the limited data problem in photoacoustic tomography and the parameter identification problems to show the effectiveness of our method.
30 November 2022 in Inverse Problems
In this paper, we focus on a class of inverse problems with Lipschitz continuous Fréchet derivatives both in Hilbert spaces and Banach spaces. The convergence and convergence rate of the inexact Newton–Landweber method (INLM) for such problems are presented under some assumptions. For the inverse problems in Hilbert spaces, we revisit the convergence result and the convergence rate of the INLM under Lipschitz condition and Hölder stability. Furthermore, the INLM for nonlinear inverse problems in Banach spaces is also considered. By using a Hölder stability corresponding to the Bregman distance, we derive the convergence property and convergence rate of the method.
27 August 2021 in Inverse Problems
We propose and analyze an accelerated homotopy perturbation iteration based on Nesterov strategy for nonlinear inverse problems in Banach spaces. The method allows to use L 1 -like and total variation-like penalty terms. Therefore, it can cope with the situation where the sought solution is sparse and piece-wise constant. Under some standard assumptions, we show the convergence and regularization properties of the method. Numerical experiments are presented to show the effectiveness as well as the acceleration effect of the proposed method.
29 May 2026 in SIAM Journal on Imaging Sciences
Stochastic gradient descent (SGD) and its variants are widely used and highly effective optimization methods in machine learning, especially for neural network training. By using a single datum or a small subset of the data, selected randomly at each iteration, SGD scales well to problem size and has been shown to be effective for solving large-scale inverse problems. In this work, we investigate SGD for solving nonlinear inverse problems in Banach spaces through the lens of iterative regularization. Under general assumptions, we prove almost sure convergence of the iterates to the minimum-distance solution and show the regularizing property in expectation under an a priori stopping rule. Further, we establish convergence rates under the conditional stability assumptions for both exact and noisy data. Numerical experiments on Schlieren tomography and electrical impedance tomography are presented to show distinct features of the method.
24 November 2022 in Journal of Inverse and Ill-posed Problems
In this paper, we propose and analyze a projected homotopy perturbation method based on sequential Bregman projections for nonlinear inverse problems in Banach spaces. To expedite convergence, the approach uses two search directions given by homotopy perturbation iteration, and the new iteration is calculated as the projection of the current iteration onto the intersection of stripes decided by above directions. The method allows to use -like penalty terms, which is significant to reconstruct sparsity solutions. Under reasonable conditions, we establish the convergence and regularization properties of the method. Finally, two parameter identification problems are presented to indicate the effectiveness of capturing the property of the sparsity solutions and the acceleration effect of the proposed method.
1 August 2025 in Physica Scripta
In this paper, we propose two versions of the homotopy perturbation iteration with inexact inner solvers (HP-IIS) for nonlinear inverse problems in both Hilbert and Banach spaces. These approaches involve the integration of an inexact inner solver to estimate the solution of the minimization problem, since solving this problem at each iteration step exactly in general is impossible in practical applications. Based on the ε-subdifferential calculus, we establish the convergence analysis of the proposed methods in both spaces. To validate the theoretical results, we implement the proposed HP-IIS methods through parameter identification inverse problems, auto-convolution inverse problems and Robin coefficient reconstruction problems. Numerical results demonstrate that while the exact solution to minimization problem remains intractable, the HP-IIS methods yield reconstructions that are quantitatively favorable. Furthermore, the proposed HP-IIS methods are still effective in scenarios where the data is corrupted by non-Gaussian noise.