一作ICML 2026Poster· Deep Learning / Large Language Models
Multilingual Safety Alignment via Representation-Space Separability
Dan Shi · Zhuowen Han · Deyi Xiong
首次获取该候选人的完整档案及 StartLux 评估共 20 积分；每位用户单独解锁，之后查看免费。主动刷新 20 积分。
实时查找电子邮箱，费用不超过20积分；未找到邮箱不扣费。
Dan Shi · Zhuowen Han · Deyi Xiong
Dan Shi · Zhuowen Han · Simon Ostermann · Renren Jin · Josef van Genabith · Deyi Xiong
Reinforcement learning (RL)-based post-training often improves the reasoning performance of large language models (LLMs) beyond the training domain, while supervised fine-tuning (SFT) frequently leads to general capabilities forgetting. However, the mechanisms underlying this contrast remain uncl...
Zhuowen Han · Lei Yang · Renren Jin · Dan Shi · Chenxi Sun · Deyi Xiong
Recently, large reasoning models have achieved impressive performance, but their lengthy reasoning processes incur substantial inference overhead. To mitigate this issue, we propose the concept of reasoning vectors, representations extracted from the model’s hidden states, which can guide the mod...
Dan Shi · Renren Jin · Zhuowen Han · Yuqi Ren · Xinwei Wu · Zhigen Li · Deyi Xiong
Despite recent advances in safety alignment, large language models (LLMs) remain highly susceptible to adversarial attacks, while the internal mechanisms behind such vulnerabilities are still poorly understood. Existing gradient-based attribution methods offer valuable interpretability for analyz...