Newest papers

Reinforcement Learning with Visually Attributable On-Policy Self-Distillation

Shiao Xie, Siyu Chen, Jianwei Lv, Xiandong Li, Yujin Wang · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.26856 · cited by 0 works (OpenAlex) · open-access PDF

Portuguese

O aprendizado de reforço visual estabelecido com recompensas verificáveis (RLVR) fornece supervisão de nível de tarefa confiável, mas oferece orientação limitada para localizar erros no raciocínio visual. A autodestilação em política (OPSD) complementa esta supervisão com correções de nível de token densas de um professor privilegiado equipado com uma visão visual centrada em evidências. No entanto, as correções resultantes podem confundir evidências visuais com mudanças não visuais induzidas por preços linguísticos, estilo de resposta ou a própria visão privilegiada. Formalizamos este fenômeno como uma contaminação de correção privilegiada de nível de token: uma correção de nível de token é contaminada quando sua direção de atualização não pode ser consistentemente atribuída à evidência visual privilegiada. A destilação direta

Machine translation (Meta M2M100); the English original follows.

Abstract

Multimodal reinforcement learning with verifiable rewards (RLVR) provides reliable task-level supervision, yet offers limited guidance for localizing errors in visual reasoning. On-policy self-distillation (OPSD) complements this supervision with dense token-level corrections from a privileged teacher equipped with an evidence-centered visual view. However, the resulting corrections may conflate visual evidence with non-visual shifts induced by linguistic priors, response style, or the privileged view itself. We formalize this phenomenon as privileged-correction contamination: a token-level correction is contaminated when its update direction cannot be consistently attributed to the privileged visual evidence. Directly distilling contaminated corrections may therefore reinforce erroneous token-level updates. To address this issue, we propose reinforcement learning with visually attributable on-policy self-distillation, an evidence-consistent paradigm that augments task-level RLVR supervision with privileged token-level corrections grounded in visual evidence. We instantiate this paradigm with Visual-Attribution Gating (VAG), which uses the verifier-derived advantage as the optimization anchor and validates each privileged correction against the contrast between evidence-present and evidence-degraded teacher views. By admitting only corrections directionally supported by this visual contrast, VAG preserves visually grounded guidance while suppressing inconsistent residuals. Experiments across six visual reasoning benchmarks show that VAG consistently outperforms GRPO and OPSD methods, establishing new state-of-the-art average performance.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Reinforcement learning, Computer science, Artificial intelligence, Contrast (vision), Human–computer interaction, Visual reasoning

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange