tracked author

Frederike Lübeck

SDPO equal second author and ETH Zürich / Max Planck Institute for Intelligent Systems machine-learning PhD candidate advised by Andreas Krause and Bernhard Schölkopf; she led the paper's code environment and test-time self-distillation design.

1 archived notes Homepage GitHub

Related Notes

按论文归档时间排序，展示该作者在本站已经出现的材料。

归档 Jul 28, 2026

Reinforcement Learning via Self Distillation

SDPO 在学生在线轨迹上用环境反馈条件化的 EMA 同模型重算下一词分布，将 Top K 分布差转成 logit 级信用；Qwen3 8B 在 LiveCodeBench v6 上达到 48.8%，高于强 GRPO 的 41.2%，但依赖模型回溯能力与反馈质量。

待审阅 2601.20802-reinforcement-learning-via-self-distillation On-Policy Distillation Credit Assignment RL Algorithm