作者位次 #4/8ACL 2026Poster· Natural Language Processing
Interpretable Safety Alignment via SAE -Constructed Low-Rank Subspace Adaptation
Dianyun Wang · Qingsen Ma · Yuhu Shang · Zhifeng Lu · Zhenbo Xu · Lechen Ning · Huijia Wu · Zhaofeng He
摘要
Safety alignment—training large language models (LLMs) to refuse harmful requests while remaining helpful—is critical for responsible deployment. Prior work established that safety behaviors are governed by low-rank structures, suggesting parameter-efficient fine-tuning (PEFT) should be well-suit...