Optimal Online Data-Driven Optimization with Multiple Time-Varying Non-Convex Objectives
Abstract
This report presents the development of optimization research methods and concepts for time-varying, real-world situations, which include: factoring in more realistic assumptions on the dynamical systems underlying the objective functions and constraints, nonconvexity of objectives, reasonability of currently existing performance measures, time-varying constraints, and situations where the true objective function is unknown. Two scenarios were considered for training of statistical models for optimization: 1.) where changes in the objective function were smooth, and 2.) where the dynamical system of objective functions was adversarial (non-predictable). The problems considered in the body of work span a large number of subfields of optimization and are summarized in the remainder of the report.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 11, 2019
- Accession Number
- AD1090195
Entities
People
- Arthur R. Calderbank
- Robert Ravier
- Vahid Tarokh
Organizations
- Duke University