14 papers
ComMem: Complementary Memory Systems for Test-Time Adaptation of Vision-Language Models
Guanglong Sun, Shuang Cui, Bo Lei +6
Test-time adaptation (TTA) of vision-language models (VLMs) is essential for their robust deployment in dynamic, real-world environments. However, existing TTA methods often adapt…
Safety Alignment as Continual Learning: Mitigating the Alignment Tax via Orthogonal Gradient Projection
Guanglong Sun, Siyuan Zhang, Liyuan Wang +3
Safety post-training can improve the harmfulness and policy compliance of Large Language Models (LLMs), but it may also reduce general utility, a phenomenon often described as the…
MePo: Meta Post-Refinement for Rehearsal-Free General Continual Learning
Guanglong Sun, Hongwei Yan, Liyuan Wang +5
To cope with uncertain changes of the external world, intelligent systems must continually learn from complex, evolving environments and respond in real time. This ability, collect…
AnyPos: Automated Task-Agnostic Actions for Bimanual Manipulation
Hengkai Tan, Yao Feng, Xinyi Mao +5
Learning generalizable manipulation policies hinges on data, yet robot manipulation data is scarce and often entangled with specific embodiments, making both cross-task and cross-p…
Dummy-Aware Weighted Attack (DAWA): Breaking the Safe Sink in Dummy Class Defenses
Yunrui Yu, Xuxiang Feng, Pengda Qin +5
Adversarial robustness evaluation faces a critical challenge as new defense paradigms emerge that can exploit limitations in existing assessment methods. This paper reveals that Du…
Why the Maximum Second Derivative of Activations Matters for Adversarial Robustness
Yunrui Yu, Hang Su, Jun Zhu
This work investigates the critical role of activation function curvature -- quantified by the maximum second derivative -- in adversarial robustness. Using the Recurs…