arXiv cs.AI· Shripad V. Deshmukh, Christian Bock, Kangrui Ruan, Abhijeet Awasthi, Vijay Lingam, Rein Houthooft, Manish Prajapat·· 1 天前AI 评分45
Projected Distillation：从贴近且正确的目标学习求解难题
Projected Distillation: On Learning to Solve Hard Problems from Close, Correct Targets
AI 导读
一篇 arXiv 论文提出 projected distillation 训练方法，将学生的失败尝试最小限度地编辑成正确目标，用于在所有尝试都失败的难题上实现近 on-policy 学习。
来源：arXiv cs.AI · arxiv.org