A statistical perspective on ill-posed inverse problems (with discussion) (Q579825): Difference between revisions

Inverse problems pertain to situations where one is interested in making inferences about a phenomenon from partial or incomplete information, and if such problems are not amenable to classical statistical estimation procedures, these are termed ill-posed. Such problems are becoming common in geophysics, medicine, meteorology, radio astronomy, reservoir engineering, tomography etc. The basic principle common to all techniques for solving inverse problems is as follows: seek a solution that is consistent both with the observed data and prior notions about the physical behaviour of the phenomenon under study. Different practical problems have led to unique strategies for implementation of this principle, such as the method of regularization, maximum entropy and quasi reversibility. In this paper the author discusses some tools for assessing the finite sample performance characteristics of an inverse algorithm. These tools are considered in the context of the generalized nonlinear regression model. Linear and nonlinear examples from medicine, meteorology and geophysics are used for illustration. The discussion on this paper is interesting and exposes several open research problems.