Dr. Quan Long from United Technologies Research Center provided an overview of his recent research on Efficient Bayesian Optimal Experimental Design for Physical Models. Optimal experimental design is the key to improve data quality in engineering. Its application on real problems lags behind mainly due to the involved computational costs. Dr. Long has developed a series of methods to accelerate the computations of the utility function (expected information gain) under rigorous error control. Specifically, he has extended the applicable domain of Laplace methods from the asymptotic posterior Gaussianity, to where the shape of the posterior is characterized by noninformative manifolds. While Laplace methods require a concentration of measure, multi-level Monte Carlo method can be used to efficiently compute the nested integral of the expected information gain with a reduction of the computational complexity, even when the randomness of data dominants the shape of the posterior distribution. The developed methodologies have been applied to various engineering problems, e.g., impedance tomography, seismic source inversion and parameter inference of combustion kinetics.
To view more details or a video of the seminar, please visit http://iase.engr.uconn.edu/events/sem-lib/sl2016/.