Convergence Rates for Learning Linear Operators from Noisy Data

Abstract

This paper studies the learning of linear operators between infinite-dimensional Hilbert spaces. The training data comprises pairs of random input vectors in a Hilbert space and their noisy images under an unknown self-adjoint linear operator. Assuming that the operator is diagonalizable in a known basis, this work solves the equivalent inverse problem of estimating the operator's eigenvalues given the data. Adopting a Bayesian approach, the theoretical analysis establishes posterior contraction rates in the infinite data limit with Gaussian priors that are not directly linked to the forward map of the inverse problem. The main results also include learning-theoretic generalization error guarantees for a wide range of distribution shifts. These convergence rates quantify the effects of data smoothness and true eigenvalue decay or growth, for compact or unbounded operators, respectively, on sample complexity. Numerical evidence supports the theory in diagonal and non-diagonal settings.

To appear in SIAM/ASA Journal on Uncertainty Quantification (JUQ); 34 pages, 5 figures, 2 tables

References in corpus (9)

• Prediction in functional linear regression
• A New Approach to Collaborative Filtering: Operator Estimation with Spectral Regularization
• Learning elliptic partial differential equations with randomized linear algebra
• Derivative-Informed Projected Neural Networks for High-Dimensional Parametric Maps Governed by PDEs
• Conditional mean embeddings as regressors - supplementary
• Bayesian inference of an uncertain generalized diffusion operator
• Deep Neural Networks Are Effective At Learning High-Dimensional Hilbert-Valued Functions From Limited Data
• Designing truncated priors for direct and inverse Bayesian problems
• Learning Schatten--von Neumann Operators

Cited by in corpus (10)

• A Mathematical Guide to Operator Learning
• Operator Learning Using Random Features: A Tool for Scientific Computing
• MCMC-Net: Accelerating Markov Chain Monte Carlo with Neural Networks for Inverse Problems
• Transfer Learning on Multi-Dimensional Data: A Novel Approach to Neural Network-Based Surrogate Modeling
• Forward and backward problems for coupled subdiffusion systems
• Fast Macroscopic Forcing Method
• Learning Partial Differential Equations in Reproducing Kernel Hilbert Spaces
• Learning a Gaussian Mixture for Sparsity Regularization in Inverse Problems
• Covariance Operator Estimation via Adaptive Thresholding
• Generalised Rank-Constrained Approximations of Hilbert-Schmidt Operators on Separable Hilbert Spaces and Applications