Machine Learning Security for LLM
Naghmeh Karimi, Department of Computer Science and Electrical Engineering.
Md Shoaib Akhter Rafi, Department of Computer Science and Electrical Engineering.
Large Language Models (LLMs) are increasingly used in critical applications, but their vulnerability to adversarial prompts, jailbreaks, prompt injection, data poisoning, and privacy attacks raises significant security concerns. This project investigates machine learning-based approaches for detecting and mitigating such threats while maintaining model performance and usability. The research focuses on analyzing LLM vulnerabilities, developing robust defense mechanisms, and evaluating their effectiveness using security and robustness metrics. The expected outcome is a practical framework for improving the security, reliability, and resilience of LLM-based systems against evolving adversarial attacks.
Synopsys Testing
Skye Jonke and Naghmeh Karimi, Department of Computer Science and Electrical Engineering.
This project will leverage Testing Synopsys scripts to adapt them for HPC computing.