Optimal tomography of dynamically evolving objects using machine learning algorithms

Abstract

In this article we introduce a systematic approach for optimal scanning of dynamically evolving objects, including cases where the dynamics is unknown. The method is specifically designed to optimize each measurement and engineer illumination patterns with the goal of reducing the uncertainty left in our estimation of the sample. Concurrently, the algorithm uses system identification techniques to develop a mathematical model for the dynamics under test based on the acquired data and it uses the model to predict changes in the distribution and optimize upcoming measurements. The theory is developed and simulations are provided to better display discussed concepts.