| title | Eliciting Weak-to-Strong Generalization with On-Policy Reverse Distillation |
|---|---|
| authors | Youngrok Park, Sangmin Bae, Hojung Jung, Jongwoo Ko, Yunseon Choi, Young Jin Kim, Pashmina Cameron, Aaron Courville, Se-Young Yun |
| date | 2026-09-08 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2609.08798 |
| affiliation | KAIST AI |
| upvotes | 68 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf code |
| abstract | Weak-to-strong generalization asks whether stronger models can learn from weaker supervisors and surpass them. This question is particularly important for successive model generations and multi-domain consolidation, where repeating frontier-scale post-training from scratch can be prohibitively expensive. Yet conventional distillation treats the weak teacher as an optimization target, potentially imposing its capacity ceiling on the student. We introduce On-Policy Reverse Distillation (OPRD), which evaluates the teacher's policy shift relative to its reference policy on student rollouts and amplifies the component of the student's verifier-driven policy gradient along that direction. By rescaling only verifier-supported updates, OPRD preserves the stationary points of policy optimization while accelerating learning beyond the teacher. In both successive model transfer and multi-teacher distillation, OPRD achieves higher performance with fewer student updates than existing RL and distillation approaches. Response-style analysis shows that OPRD students remain closer to models trained with verifier-based RL alone than to their weak teachers, suggesting that teacher guidance accelerates rather than redirects the student's own optimization. Results in conventional strong-to-weak distillation further demonstrate that OPRD effectively combines verifier-driven policy optimization with teacher guidance regardless of capacity ordering. |
cite
@misc{park2026eliciting,
title={Eliciting Weak-to-Strong Generalization with On-Policy Reverse Distillation},
author={Youngrok Park and Sangmin Bae and Hojung Jung and Jongwoo Ko and Yunseon Choi and Young Jin Kim and Pashmina Cameron and Aaron Courville and Se-Young Yun},
year={2026},
eprint={2609.08798},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2609.08798}
}