Zeroth- and First-Order Methods for Bilevel Optimization

Abstract

Bilevel optimization (BO), also known as two-level optimization, is a critical branch of mathematical optimization. It involves two distinct optimization problems- an upper-level problem representing the decisions of a leader, and a lower-level problem representing the decisions of a follower. These problems are inherently connected, with solutions from one level influencing the other, creating a hierarchical and dynamic structure of decision-making. Numerous decision problems arising in DoD scenarios, such as resource allocation, logistics, radar system deployment, and engineering design, often exhibit a hierarchical structure and can be addressed as BO problems. However, solving BO is generally challenging due to its nested nature and the simultaneous consideration of multiple objectives and constraints. Additionally, it frequently involves sophisticated objective functions and constraints at both levels. Moreover, BO problems are often large-scale, especially in the big data era, characterized by objective functions and constraints involving a plethora of functions. To address these challenges, the research will develop novel computational methods to solve a broad spectrum of important and challenging BO problems.

Document Details

Document Type
DoD Grant Award
Publication Date
Feb 06, 2025
Source ID
FA95502410343

Entities

People

  • Zhaosong Lu

Organizations

  • Air Force Office of Scientific Research
  • Regents of the University of Minnesota
  • United States Air Force

Tags

Readers

  • Marine Propulsion Engineering and Naval Architecture
  • Operations Research
  • Systems Analysis and Design