Applied Mathematics and Statistics Seminar by Ruhui Jin:Data Selection for Reliable Learning: Probabilistic Experimental Design

Speaker: Ruhui Jin, Department of Applied Mathematics and Statistics, Johns Hopkins University

 

 

Learning the underlying mechanisms of physical and biological systems from indirect observations has long been a central challenge in applied mathematics. One caution, however, is that practical measurements are inevitably imperfect for various reasons. Together with limited experimental and simulation resources, it motivates a key question: How can we develop mathematically sound strategies for selecting valuable data to support reliable and effective learning?

Rather than taking arbitrary data, we turn to optimal experimental design, a statistical framework for guiding data acquisition, that has gained renewed interest in computational inverse problems. Through examples such as optimal sensor placement and detection timing, we explore the probabilistic aspects of selecting informative measurements. We conclude by discussing how randomized algorithms can improve the computational efficiency of experimental design, but at the expense of some degree of optimality.