DesLearn- A Generalised Framework for Seamless Integration of Machine Learning and Discrete-Event Simulation
Abstract
The field of generative AI has witnessed remarkable advances in recent times, particularly in the generation of high-quality texts, images, and speech. At the heart of these advancements is the use of Large Language Models (LLMs). While this technology paves the way for numerous beneficial applications, it also presents potential for malicious misuse. In this project, PI will center the focus on ensuring a robust alignment between LLM outputs and human values. Initially, PI will introduce a novel fine-tuning method designed to enable safe fine-tuning of LLMs, ensuring they align with desired properties. Subsequently, they will craft a resilient algorithm to fortify LLMs against adversarial attacks, making it difficult for them to be deceived. Through this combined effort, they aim to provide a comprehensive method to align LLMs closely with human values.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 05, 2025
- Source ID
- FA23862414017
Entities
People
- Paul Corry
Organizations
- Air Force Office of Scientific Research
- Queensland University of Technology
- United States Air Force