Guardado en:
Detalles Bibliográficos
Autores principales: Zhang, Yuwei, Li, Sha, Yu, Changlong, Lu, Qin, Jin, Shuowei, Dong, Chengyu, Liu, Haoran, Hong, Ilgee, Li, Xintong, Shi, Zhenyu, Yin, Bing, Shang, Jingbo
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2605.12741
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866911677331537920
author Zhang, Yuwei
Li, Sha
Yu, Changlong
Lu, Qin
Jin, Shuowei
Dong, Chengyu
Liu, Haoran
Hong, Ilgee
Li, Xintong
Shi, Zhenyu
Yin, Bing
Shang, Jingbo
author_facet Zhang, Yuwei
Li, Sha
Yu, Changlong
Lu, Qin
Jin, Shuowei
Dong, Chengyu
Liu, Haoran
Hong, Ilgee
Li, Xintong
Shi, Zhenyu
Yin, Bing
Shang, Jingbo
contents Enabling Large Language Models (LLMs) to continuously improve from environmental interactions is a central challenge in post-training. While on-policy self-distillation offers a promising paradigm, existing methods predominantly treat environmental feedback as a passive conditioning signal. Consequently, they heavily rely on successful demonstrations and struggle to learn in rare-success regimes. To bridge this gap, we introduce Reflection-Enhanced Self-Distillation (RESD), a framework that transforms raw failure feedback into an active source of corrective supervision. Instead of passively appending feedback, RESD interprets failed trajectories by generating retrospective reflections to diagnose local errors, and curates a persistent global playbook to preserve reusable lessons across training steps. The enriched context enables the self-teacher to provide actionable token-level supervision even in the absence of successful rollouts. Empirical evaluations on multiple continual learning tasks demonstrate that RESD substantially outperforms standard self-distillation baselines. Furthermore, RESD achieves significantly faster early-stage improvement than GRPO with $8\times$ samples using only a single rollout per prompt, highlighting its superior interaction efficiency.
format Preprint
id arxiv_https___arxiv_org_abs_2605_12741
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Learning with Rare Success but Rich Feedback via Reflection-Enhanced Self-Distillation
Zhang, Yuwei
Li, Sha
Yu, Changlong
Lu, Qin
Jin, Shuowei
Dong, Chengyu
Liu, Haoran
Hong, Ilgee
Li, Xintong
Shi, Zhenyu
Yin, Bing
Shang, Jingbo
Machine Learning
Enabling Large Language Models (LLMs) to continuously improve from environmental interactions is a central challenge in post-training. While on-policy self-distillation offers a promising paradigm, existing methods predominantly treat environmental feedback as a passive conditioning signal. Consequently, they heavily rely on successful demonstrations and struggle to learn in rare-success regimes. To bridge this gap, we introduce Reflection-Enhanced Self-Distillation (RESD), a framework that transforms raw failure feedback into an active source of corrective supervision. Instead of passively appending feedback, RESD interprets failed trajectories by generating retrospective reflections to diagnose local errors, and curates a persistent global playbook to preserve reusable lessons across training steps. The enriched context enables the self-teacher to provide actionable token-level supervision even in the absence of successful rollouts. Empirical evaluations on multiple continual learning tasks demonstrate that RESD substantially outperforms standard self-distillation baselines. Furthermore, RESD achieves significantly faster early-stage improvement than GRPO with $8\times$ samples using only a single rollout per prompt, highlighting its superior interaction efficiency.
title Learning with Rare Success but Rich Feedback via Reflection-Enhanced Self-Distillation
topic Machine Learning
url https://arxiv.org/abs/2605.12741