Title:A New Global Stochastic Search Approach for Inverse Problems: Application to Ultrasound Modulated Optical Tomography

Abstract:A global stochastic search method, which is strictly derivative-free yet directed through a gain-based additive update term, is proposed and applied to the inverse problem of ultrasound modulated optical tomography (UMOT). The additive update term results from a change of measure that is effected to drive appropriately derived observation-prediction error terms to zero-mean Wiener martingales, a characterization made possible following a representation of the discretized parameter vector to be reconstructed as a diffusion vector process. The component processes are defined as evolving with respect to an iteration parameter, leading to a recursive prediction-update algorithm to implement the stochastic search. Numerical evidence based on solutions to the UMOT problem indicates substantive improvement in the reconstruction accuracy through the proposed method vis-a-vis the popular Gauss-Newton approach.