paper-with-me

홈 › Papers

Where to Look Matters: On-Policy Self-Distillation for Long-Video Understanding

2026-08-26 · Kaishen Wang, Dongdi Zhao, Yijun Liang, Dingqiang Ye, Ruibo Chen, Heng Huang, Di Fu arxiv

Vision-language models (VLMs) have made substantial progress in long-video understanding, with standard backbone models typically answering questions from frames sampled across the full video. However, as videos become longer, the full-video context inevitably contains more question-irrelevant temporal content, which can distract the model from the evidence needed to answer a specific question. We empirically find that focusing the visual input on short annotated clue intervals containing question-relevant evidence consistently improves prediction accuracy across model scales compared with using the corresponding full videos, while requiring fewer input frames. Based on this finding, we introduce Clue-OPSD, a clue-privileged on-policy self-distillation framework for long-video understanding. During training, a full-video student learns from a self-teacher conditioned on the corresponding clue interval by aligning their next-token distributions along student-generated trajectories. Clue-OPSD thus uses clue intervals as privileged supervision without relying on ground-truth answer labels, while requiring no clue annotations or additional modules at inference time. Extensive experiments across multiple long-video understanding benchmarks and Qwen3.5 model scales demonstrate consistent improvements over the corresponding backbone models and strong performance against supervised post-training baselines.

📄 PDF Abstract BibTeX arXiv:2608.25356

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

Self-Supervised Visual On-Policy Distillation

2026-08-14 · Yijiang Li, Yijun Liang, Yunjie Tian, Bingyang Wang 외 arxiv

Visual on-policy distillation relies heavily on an informative teacher-student asymmetry, through either a larger, stronger teacher or privileged supervision, such as reference answers or ground-truth regions of interest…

Localizing Credit at the Divergence: Path-Conditioned Self-Distillation for LLM Reasoning

2026-06-14 · Yu Li, Shu Hong, Tian Lan arxiv

Reinforcement learning from verifiable rewards assigns a single scalar to each rollout, leaving token-level credit assignment underspecified in long reasoning traces. On-policy self-distillation addresses this by letting…

Reinforcement Learning

OISD: On-Policy Internal Self-Distillation of Language Models

2026-05-27 · Xinyu Liu, Darryl Cherian Jacob, Yang Zhou, Jindong Wang 외 arxiv

Recent reinforcement learning (RL) post-training approaches primarily optimize the final output policy using sparse outcome-level rewards, while largely overlooking predictive signals encoded in intermediate representati…

Reinforcement LearningMathematical Reasoning

TRACE: Distilling Where It Matters via Token-Routed Self On-Policy Alignment

2026-05-11 · Jiaxuan Wang, Xuan Ouyang, Zhiyu Chen, Yulan Hu 외 arxiv

On-policy self-distillation (self-OPD) densifies reinforcement learning with verifiable rewards (RLVR) by letting a policy teach itself under privileged context. We find that when this guidance spans the full response, a…

Reinforcement Learning

Embarrassingly Simple Self-Distillation Improves Code Generation

2026-04-01 · Ruixiang Zhang, Richard He Bai, Huangjie Zheng, Navdeep Jaitly 외 arxiv

Can a large language model (LLM) improve at code generation using only its own raw outputs, without a verifier, a teacher model, or reinforcement learning? We answer in the affirmative with simple self-distillation (SSD)…

Reinforcement LearningCode Generation