CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Statistical filtering for optimization under uncertainty Authors:  Vivak Patel - University of Wisconsin -- Madison (United States) [presenting]
Abstract: Across many disciplines, inference or decision tasks are formulated as optimizing objective functions involving expectations. While there are several paradigms for addressing such problems, such as Bayesian optimization or stochastic gradient methods, these paradigms are either computationally impractical or are too underdeveloped for real applications. Thus, there is still a need for practical optimization methods for these optimization problems. We introduce a novel paradigm that addresses this concern. Our paradigm leverages statistical filters to generate computationally practical subproblems, which can then be solved by mature, deterministic optimization methods. The resulting algorithms perform surprisingly well against state-of-the-art approaches, which we demonstrate on a handful of problems from a number of application areas.