Mixed-integer nonlinear programming- Unleashing the full potential of relaxations

Abstract

Mixed-integer optimization techniques have been successfully deployed to tackle decision-making problems in machine learning, robotics, logistics and policy making. However, modern optimization problems arising in highly uncertain environments, or in the presence of multiple agents and adversaries, call for the introduction of nonlinearities and the development of increasingly sophisticated methods. A fundamental technique, convexification, transforms highly complex landscapes that arise in mixed-integer optimization problems into simpler but equivalent forms, ensuring that optimal solutions can be easily spotted. Unfortunately, solvers for mixed-integer problems with nonlinearities, based on the popular branch-and-bound algorithm, struggle to incorporate several convexification results that have been derived in the literature, severely limiting their potential to tackle challenging, large-scale problems. The proposed research will develop novel methods to exploit convexification results, allowing for a natural integration of different convex relaxations. The methods will exploit the strength and quality of the more sophisticated relaxations, and at the same time maintain the tractability of the simpler relaxations. Consequently, the proposed techniques are fully compatible, and will significantly enhance, existing methods. The project will also contribute to the literature in convexification, by developing new theory, producing novel convex relaxations, and proposing new paradigms to tackle challenging nonlinear problems. If successful, the project will lead to significant advances in solving critical problems arising in decision-making with uncertainty, control and machine learning.

Document Details

Document Type
DoD Grant Award
Publication Date
Feb 06, 2025
Source ID
FA95502410086

Entities

People

  • Andres Gomez Escobar

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • University of Southern California

Tags

Readers

  • Operations Research
  • Systems Analysis and Design

Technology Areas

  • AI & ML
  • AI & ML - Machine Learning Algorithms
  • AI & ML - Neural Networks
  • Autonomy
  • Autonomy - Autonomous System Control