A Morse-Theoretic Approach to Non-Convex Optimization

Abstract

Non-convex optimization is becoming the foundation for a wide range of applications in modern engineering. Examples include deep learning, signal processing, robotics, autonomy, and numerous others. The class of non-convex problems is broader than the class of convex problems, and the extensive impact of convex optimization suggests the theory of non-convex optimization is poised to match or surpass that impact. However, non-convex optimization remains poorly understood. Many standard analyses for convex problems do not carry over to the non-convex setting, and the analysis of non-convex optimization algorithms is often ad-hoc or heuristic (if done at all). Stronger theoretical guarantees are not only desirable in academic and industrial research, but they are also a pre-requisite for the deployment of neural networks and other technologies in realworld safety-critical systems. To enable novel, rigorous performance guarantees, this YIP project will establish new connections between Morse theory and non-convex optimization. Optimization algorithms will be studied as dynamical systems, and convergence to minimizers will be analyzed as stability.

Document Details

Document Type
DoD Grant Award
Publication Date
Feb 29, 2024
Source ID
FA95502310120

Entities

People

  • Matthew Hale

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • University of Florida

Tags

Readers

  • Agent-Based Social Robotics and Mobile-Assisted Learning in Virtual Environments.
  • Neural Network Machine Learning.
  • Operations Research

Technology Areas

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