Developing Novel Triggers for Backdoor Attacks on LLMs | Collin Francel | RRS Live 2025
Автор: Randall Research Scholars Program
Загружено: 2026-02-23
Просмотров: 7
Описание:
Collin Francel is a Junior Data Science, Mathematics major in the Randall Research Scholars Program (RRSP). Their research project presentation, "Developing Novel Triggers for Backdoor Attacks on LLMs," was completed under the advisement of Sayanton Dibbo from the Department of Computer Science.
Project Description:
Security of large language models poses a significant challenge, as they are vulnerable to a large range of attacks. If attackers have direct access to the model, they may use fine tuning attacks to inject malicious information into the model, and distribute the compromised model to unsuspecting users on platforms like HuggingFace. This work examines new possible attack vectors, focusing specifically on developing novel triggers that the model is fine tuned to associate with negative behaviors.
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
