Deep neural network yields regularization for ill-posed inverse problems
Provides a theoretical framework for regularization in deep neural networks for ill-posed inverse problems, addressing a known bottleneck in applying DNNs to such problems.
This paper extends architecture-based regularization from shallow to deep neural networks for ill-posed inverse problems, proposing two discrepancy-principle-driven expanding DNN algorithms. Theoretical results include finite termination, convergence as noise vanishes, and explicit bounds on network complexity, with numerical experiments validating the approach.
This paper studies the regularization of ill-posed inverse problems by deep neural networks (DNNs). We extend architecture-based regularization from shallow networks to deep models by developing a deterministic framework in which the admissible network class is enlarged adaptively and the resulting architecture complexity acts as the regularization mechanism. We propose two discrepancy-principle-driven expanding DNN algorithms to treat the cases where an explicit parameter-radius bound is available and unavailable, respectively. For both algorithms, we prove the finite termination of the adaptive expansion procedure and the convergence of the regularized solutions as the noise level vanishes. In addition, we derive explicit asymptotic bounds on the terminal network architecture, thereby quantifying how the required network complexity scales with the noise level. Numerical experiments on several representative linear and non-linear inverse problems support the theoretical findings and illustrate the practical usefulness of the proposed framework.