A Unified Perspective on Robustness, Collaboration, and Fairness in AI
Abstract
This project contributes a unified perspective on the design of machine learning systems for use in strategic and societal settings. The primary objective of this proposal is to develop a uni-fied theoretical foundation for addressing robustness, multi-agent collaboration, and fairness in AI systems. Building on the recently introduced multiobjective and multidistribution learning frame-works, the main thrusts of the proposal address statistical, computational, and strategic challenges that arise in the addressing complexmultiobjective learning systems. As a secondary objective, this proposal explores how lessons learned from this mathematical foundation can guide a con-ceptual approach to addressing robustnessness of real-life complex systems such as large language models.Robustness of autonomous agents and multi-agent collaborations are a crucial part of the naval landscape of the future, with examples of naval applications including the design of intelligent autonomous agents thatperform well in uncertain physical environments, the design of systems that effectively handle adverse and malicious interventions, enhancing situational awareness using collaborations between decentralized drones and probes, and ensuring the reliability of generative AI methods used for multi-model and multi-lingual communication.Approved for Public Release
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 08, 2024
- Source ID
- N000142412159
Entities
People
- Nika Haghtalab
Organizations
- Office of Naval Research
- United States Navy
- University of California Regents