Skill-Decoupled On-Policy Expert Distillation for Unified Vision-Language Models
Summary
该论文提出 SCOPED（Skill-Decoupled On-Policy Expert Distillation），通过训练能力专属专家并以奖励引导反思提取技能，在在线蒸馏中将程序性知识内化到统一的视觉-语言模型中，从而避免跨能力梯度干扰，并在八个多模态基准上超越骨干模型和混合能力 RL。
View Cached Full Text
Cached at: 10/16/26, 07:04 AM
Paper page - Skill-Decoupled On-Policy Expert Distillation for Unified Vision-Language Models
Source: https://huggingface.co/papers/2610.22217
Abstract
SCOPED trains separate experts for distinct capabilities, extracts skills via reward-guided reflection, and distills procedural knowledge into a unified vision-language model without exposing explicit skills at inference, reducing cross-capability interference.
Reinforcement learning(RL) can elicit specialized visual capabilities invision-language models(VLMs), yet jointly optimizing heterogeneous objectives often introducescross-capability gradient conflicts. Training experts independently mitigates this interference, but direct reward optimization may entangle specialized knowledge with reward-hacking biases. This paper proposesSkill-DecoupledOn-PolicyExpert Distillation(SCOPED), which trains capability-specific experts and extracts explicit skills throughreward-guided reflection. Duringon-policy distillation, these skills guide expert supervision while remaining unavailable to the unified student, encouraging it to internalize procedural knowledge without explicit skills at inference time. Across eightmultimodal benchmarks, SCOPED consistently outperforms the backbone and mixed-capability RL, demonstrating the effectiveness ofskill-decoupledon-policyexpert distillationfor integrating independently acquired capabilities. Further analyses reveal interference during joint capability optimization and suggest that skill augmentation promotes more consistent expert trajectories, greater attention to task-relevant visual evidence, and better preservation of specialized expertise in the unified student. All code and data are available at https://github.com/NEUIR/SCOPED.
View arXiv pageView PDFGitHub2Add to collection
Get this paper in your agent:
hf papers read 2610\.22217
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2610.22217 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2610.22217 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2610.22217 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
OPID: On-Policy Skill Distillation for Agentic Reinforcement Learning
OPID proposes an on-policy skill distillation framework that extracts dense hindsight supervision from completed trajectories, combining outcome-based RL with token-level self-distillation to improve language agent training efficiency and performance on multi-turn tasks.
SCOPD: Sparse-Context On-Policy Self-Distillation for Efficient Vision-Language Models
SCOPD introduces a sparse-context on-policy self-distillation framework that enhances the efficiency of vision-language models by improving the use of pruned visual tokens, achieving higher performance retention without additional inference cost.
COLLEAGUE.SKILL: Automated AI Skill Generation via Expert Knowledge Distillation
This paper presents COLLEAGUE.SKILL, an open-source system for automatically distilling person-grounded AI skills from heterogeneous traces into inspectable, correctable, and portable skill packages, enabling LLM agents to carry bounded representations of human expertise and interaction style.
Self-Boosting Vision-Language Models with Noisy Student On-Policy Self-Distillation
Proposes NOPD, a self-distillation method that improves vision-language models without external supervision by leveraging prediction discrepancies between clean and corrupted inputs. Achieves significant gains on visual reasoning tasks, matching or exceeding RL and distillation from external models.
SKILL-KD: Contrastive Skill Distillation for LLM Agents
SKILL-KD is a contrastive skill distillation framework that improves LLM agents by distilling actionable discrepancies between teacher and student trajectories into textual skill patches, with drift-aware consolidation to iteratively refine skills.