On the asymptotical regularization for linear inverse problems in presence of white noise

Abstract

We interpret steady linear statistical inverse problems as artificial dynamic systems with white noise and introduce a stochastic differential equation (SDE) system where the inverse of the ending time $T$ naturally plays the role of the squared noise level. The time-continuous framework then allows us to apply classical methods from data assimilation, namely the Kalman-Bucy filter and 3DVAR, and to analyze their behaviour as a regularization method for the original problem. Such treatment offers some connections to the famous asymptotical regularization method, which has not yet been analyzed in the context of random noise. We derive error bounds for both methods in terms of the mean-squared error under standard assumptions and discuss commonalities and differences between both approaches. If an additional tuning parameter $\alpha$ for the initial covariance is chosen appropriately in terms of the ending time $T$, one of the proposed methods gains order optimality. Our results extend theoretical findings in the discrete setting given in the recent paper Iglesias et al. (2017). Numerical examples confirm our theoretical results.

References in corpus (8)

• The Ensemble Kalman Filter for Inverse Problems
• Data Assimilation: A Mathematical Introduction
• Characterizations of variational source conditions, converse results, and maxisets of spectral regularization methods
• Well Posedness and Convergence Analysis of the Ensemble Kalman Inversion
• Beyond the Bakushinskii veto: Regularising linear inverse problems without knowing the noise distribution
• Generalized modes in Bayesian inverse problems
• On parameter identification in stochastic differential equations by penalized maximum likelihood
• Empirical Risk Minimization as Parameter Choice Rule for General Linear Regularization Methods