Optimization and Learning with Contextual Risk Models
Abstract
The project focuses on stochastic optimization problems which involve risk evaluation and management. Specifically, the focus is on problems in which additional information, called the context, become available before decisions are made. In such situations, the challenge is to find a decision rule rather than one decision that is best overall. Problems of this nature pose a significant theoretical and computational challenge because they require the analysis and optimization of the context-dependent risk and the balancing of risk across the contexts. In the project, several contextual risk-averse optimization problems will be analyzed, involving problems with conditional measures of risk and with conditional distributional constraints. The research plan includes the mathematical analysis of these problems and the development of effective numerical methods for their solutions. The focus will be on problems involving non-differentiable and non-convex functions. Special attention will be paid to stochastic learning methods, which will optimize the system by simulation or in an online fashion. The proposed modeling paradigm and solution methodology will have broad applicability in industry and science sectors and is relevant to military operations. Risk is fundamental for reliability, disaster management, logistics, security, and many other areas. The idea to consider contextual risk will achieve a better understanding of the nature of risk and to manage it more precisely and effectively.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 06, 2025
- Source ID
- FA95502410284
Entities
People
- Andrzej Piotr Ruszczyński
Organizations
- Air Force Office of Scientific Research
- Rutgers University
- United States Air Force