Department of Statistics
Environmental Seminar Series
North Carolina State University


Arvind Saibaba

FROM Mathematics  

FROM NC State University

Generalized Hybrid Iterative Methods for Large-Scale Inverse Problems


Inverse problems use physical measurements to make inferences about parameters. We adopt the Bayesian approach for solving inverse problems; however, this is computationally challenging, especially for large-scale problems. In this talk, I will discuss recently developed iterative methods for computing solutions to inverse problems, in which the parameter-to-observable map is linear.  These methods are particularly useful for large-scale problems where covariance kernels, e.g., those from the Mat'ern class,  are defined on irregular grids and the resulting covariance matrices are only available via matrix-vector multiplication.  Our approach shares many benefits of standard hybrid methods such as avoiding semi-convergence and automatically estimating the regularization parameter. Numerical examples from seismic travel-time tomography and super-resolution, demonstrate the effectiveness of the described approaches. I will also discuss some ongoing work on efficiently computing uncertainty estimates by sampling from the posterior distribution.

Joint work with Julianne Chung.

Thursday, 23 March, 2017
4:30-5:30 pm

1108 SAS Hall

Refreshments will be served in the 5th floor commons at 3:45 pm.
NOTE: No food or drink is allowed in any of the classrooms in SAS Hall.