[
    {
        "id": "osp-15870",
        "type": "article-journal",
        "title": "MeSD: Multi-Evidence Self-Distillation for VideoLLM",
        "author": [
            {
                "family": "Zhu",
                "given": "Weijie"
            },
            {
                "family": "Fang",
                "given": "Han"
            },
            {
                "family": "Fu",
                "given": "Hanyu"
            },
            {
                "family": "Zhang",
                "given": "Yuzhe"
            },
            {
                "family": "Wei",
                "given": "Xin"
            },
            {
                "family": "Pan",
                "given": "Zhaoyan"
            },
            {
                "family": "Liu",
                "given": "Feiran"
            },
            {
                "family": "Jin",
                "given": "Xunjie"
            },
            {
                "family": "Sun",
                "given": "Hongbo"
            },
            {
                "family": "Lin",
                "given": "Zhiyu"
            },
            {
                "family": "Gao",
                "given": "Tianyi"
            },
            {
                "family": "Ding",
                "given": "Tianyi"
            },
            {
                "family": "Yuan",
                "given": "Ye"
            },
            {
                "family": "He",
                "given": "Zhongjiang"
            },
            {
                "family": "Sun",
                "given": "Hao"
            },
            {
                "family": "Wu",
                "given": "Zhiheng"
            }
        ],
        "URL": "https://omanscience.com/en/articles/mesd-multi-evidence-self-distillation-for-videollm",
        "language": "en",
        "issued": {
            "date-parts": [
                [
                    2026
                ]
            ]
        },
        "abstract": "While reinforcement learning with verifiable rewards provides reliable outcome supervision for VideoLLMs, sequence-level rewards offer limited token-level guidance. On-policy self-distillation addresses this limitation by conditioning a self-teacher on privileged information to provide dense token-level supervision. However, aggregating heterogeneous evidence within a single teacher context obscures cross-evidence agreement and conflict. A further challenge lies in determining whether teacher guidance should refine reward-based updates or provide corrective supervision for failed trajectories. To address these issues, we propose MeSD, a multi-evidence self-distillation framework for VideoLLMs. MeSD constructs three evidence-conditioned teachers with shared parameters, using the ground-truth answer as a common semantic context while separately incorporating temporal and spatial evidence. Given the same student-generated prefixes, MeSD evaluates evidence-specific preferences relative to the Answer Teacher and fuses teacher-common preferences with gated teacher-specific residuals. Furthermore, MeSD introduces Verification-Guided Optimization to classify trajectories as Success, Failure, or Indeterminate. For Success and Indeterminate trajectories, MeSD refines token-level advantage magnitudes while preserving reward-derived signs. For verified failure trajectories that contain the required evidence, MeSD applies failure-conditioned distillation, using reverse-KL correction toward the fused distribution. Experiments on multiple video benchmarks demonstrate consistent gains over reinforcement learning and self-distillation baselines."
    }
]