“Enhancing statistical performances via probabilistic optimizations” by Dr. Henry Lam
Dr. Henry Lam
Assistant Professor of Industrial & Operations Engineering
College of Engineering
University of Michigan
I will talk about the use of probabilistic optimizations (i.e. optimizations posited over probability distributions) to remedy some significant issues of current statistical methods in two problems. The first is extreme event analysis, where the goal is to estimate tail risk quantities given scarce tail data. I will talk about a tailconvexity‐based optimization that alleviates the model biases exhibited by conventional approaches using extreme value theory. The second problem is stochastically constrained optimization, where the goal is to obtain a decision that is confidently feasible with only limited observations on the stochastic resources. I will talk about a divergence‐based reformulation of the constraint that attenuates the over‐conservativeness inherent in common methods such as sample average approximation.