arXiv cs.AI· Linfei Ji (Huazhong University of Science and Technology), Jiayun Li (Tsinghua University), Haozhao Wang (Huazhong University of Science and Technology), Siyu Jiang (Huazhong University of Science and Technology), Zehao Wang (Harbin Institute of Technology), Longtao Xiao (Huazhong University of Science and Technology), Yu-Jie Yuan (Huawei Technologies Ltd), Ruixuan Li (Huazhong University of Science and Technology)·· 11 天前AI 评分34
TeachBack：面向在线自蒸馏的双向偏好优化框架
TeachBack: Learning to Teach via Bidirectional Preference Optimization for On-Policy Self-Distillation
AI 导读
研究人员提出 TeachBack 框架，通过双向偏好优化解决在线自蒸馏（OPSD）中特权教师模型过度依赖参考答案、缺少推导与纠错过程的问题。该框架通过调控提示词中参考推理后缀的长度与风格促使教师自主解题，并利用成对偏好目标实现教师与学生轨迹的双向优化互促。实验表明，TeachBack 在数学和逻辑推理任务上的表现持续优于现有在线自蒸馏方法及相关基准。
来源：arXiv cs.AI · arxiv.org