- This event has passed.
Research Challenges in Computing
November 15 @ 11:00 am - 12:00 pm
November 15, 2024
11:00pm
3147 MEB
https://utah.zoom.us/j/99963204158?pwd=tEpfDa4bQrzT7echP32lnt3LbY8fyP.1
Meeting ID: 999 6320 4158
Passcode: 595106
___________________________________________________________________
Guanhong Tao
Assistant Professor
The Curse of the Auto-regressive Mechanism in Large Language Models
Abstract: Large language models (LLMs) are an emerging machine learning technique with applications across various sectors, such as code generation, route planning, and threat detection. However, like conventional machine learning algorithms, LLMs also have security vulnerabilities. In this talk, I will introduce our recent work on exploring the unique properties (the auto-regressive mechanism) of LLMs to identify and mitigate these vulnerabilities. The first part will focus on jailbreaking LLMs, where we aim to induce toxic responses without crafting specific input prompts. The second part will cover current backdoor attacks on LLMs and our approaches to defending against such adversaries.
___________________________________________________________________
Jun Xu
Assistant Professor
Autonomous Software Security with 2024’s AI: Some Observations as a DARPA AIxCC Player
Abstract: In this talk, I would like to share some observations about society’s progress in autonomous software security, from the perspective of a DARPA AIxCC (AI Cyber Challenge) player. The talk will consist of two parts. In the first part, I will introduce the definition and setup of autonomous software security adopted by AIxCC, explain the participation and process of the challenge, and summarize the collective results. In the second part, I will elaborate on what techniques work better for autonomous software security and where and how today’s AI may help.