Krylov methods for inverse problems have the nice property that regularization can be decided dynamically. However, this typically requires that the entire Krylov space is kept in memory, which is problematic for large problems that do not converge quickly. We propose strategies for truncating the search space while maintaining the possibility of dynamic regularization (for various regularization methods). In addition, these strategies have advantages if a sequence of related regularized solves is required.

This presentation is part of Minisymposium “MS8 - Krylov Methods in Imaging: Inverse Problems, Data Assimilation, and Uncertainty Quantification (2 parts)”
organized by: Arvind Saibaba (North Carolina State University) , Julianne Chung (Virginia Tech) , Eric de Sturler (Virginia Tech) .

Authors:

Eric de Sturler (Virginia Tech)

Julianne Chung (Virginia Tech)

Keywords:

computed tomography, image deblurring, image reconstruction, numerical linear algebra