References & Citations
Mathematics > Numerical Analysis
Title: An adaptive RKHS regularization for Fredholm integral equations
(Submitted on 24 Mar 2023 (this version), latest version 4 Dec 2023 (v3))
Abstract: Regularization is a long-standing challenge for ill-posed linear inverse problems, and a prototype is the Fredholm integral equation of the first kind. We introduce a practical RKHS regularization algorithm adaptive to the discrete noisy measurement data and the underlying linear operator. This RKHS arises naturally in a variational approach, and its closure is the function space in which we can identify the true solution. Furthermore, we prove and numerically demonstrate that the RKHS-regularized estimator has a mean-square error converging linearly as the noise scale decays. In contrast, the commonly-used $L^2$-regularized estimator has a flat mean-square error.
Submission history
From: Fei Lu [view email][v1] Fri, 24 Mar 2023 01:43:42 GMT (130kb,D)
[v2] Wed, 29 Mar 2023 16:31:10 GMT (130kb,D)
[v3] Mon, 4 Dec 2023 22:27:30 GMT (690kb,D)
Link back to: arXiv, form interface, contact.