3 papers
cs.LG2026
RUB: Evaluating Residual Knowledge in Unlearned Models
Hao Xuan, Xingyu Li
Machine Unlearning (MUL) has emerged as a key mechanism for privacy protection and content regulation, yet current techniques often fail to guarantee the complete removal of sensit…
cs.CR2026
TrajGuard: Streaming Hidden-state Trajectory Detection for Decoding-time Jailbreak Defense
Cheng Liu, Xiaolei Liu, Xingyu Li +2
Existing jailbreak defense paradigms primarily rely on static detection of prompts, outputs, or internal states, often neglecting the dynamic evolution of risk during decoding. Thi…
cs.LG2025
Exploring the Impact of Temperature Scaling in Softmax for Classification and Adversarial Robustness
Hao Xuan, Bokai Yang, Xingyu Li
The softmax function is a fundamental component in deep learning. This study delves into the often-overlooked parameter within the softmax function, known as "temperature," providi…