超越感知捷径：面向轻量级多模态大模型可泛化视频推理的因果启发去偏优化
Beyond Perceptual Shortcuts: Causal-Inspired Debiasing Optimization for Generalizable Video Reasoning in Lightweight MLLMs
对做轻量级视频 MLLM 的研究者，这篇给出了一个可复用的去偏训练范式：先训练一个专门拟合数据偏置的「偏置模型」，再用带排斥项的 CDPO 目标把主模型推离捷径逻辑，可在不做 SFT、仅用 1% 数据做 RL 的条件下提升视频推理泛化性。
原文依据 employing an innovative repulsive objective to actively push it away from the bias model's flawed logic
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论尽管强化学习（RL）显著提升了大语言多模态模型（MLLM）的推理能力，但其对边缘部署所必需的轻量级模型效果仍然有限。为应对这一问题，我们借助因果分析与实验揭示了感知偏置这一潜在现象，表明基于 RL 的微调会迫使轻量级模型优先采用由数据偏置诱导的感知捷径，而非发展出真正的推理能力。受此启发，我们提出 VideoThinker，一个通过两阶段去偏过程在轻量级模型中培养稳健推理能力的因果启发框架。首先，偏置感知训练阶段构建一个专门的「偏置模型」以体现这些捷径行为。随后，因果去偏策略优化（CDPO）算法对主模型进行微调，采用一种创新的排斥目标，在主动将其推离偏置模型错误逻辑的同时，将其拉向正确且可泛化的解。我们的模型 VideoThinker-R1 在视频推理效率上确立了新的最优。在同规模比较中，它无需监督微调（SFT），且仅使用 1% 的训练数据用于 RL，就在广泛使用的基准上以 3.2% 的平均增益超越 VideoRFT-3B，并在 VideoMME 上领先 7%。在跨规模比较中，它在多个基准上优于更大的 Video-UTR-7B 模型，包括在 MVBench 上取得 2.1% 的增益、在 TempCompass 上取得 3.8% 的增益。代码见 https://github.com/falonss703/VideoThinker。
英文原文摘要 1615 字符
Although reinforcement learning (RL) has significantly advanced reasoning capabilities in large multimodal language models (MLLMs), its efficacy remains limited for lightweight models essential for edge deployments. To address this issue, we leverage causal analysis and experiment to reveal the underlying phenomenon of perceptual bias, demonstrating that RL-based fine-tuning compels lightweight models to preferentially adopt perceptual shortcuts induced by data biases, rather than developing genuine reasoning abilities. Motivated by this insight, we propose VideoThinker, a causal-inspired framework that cultivates robust reasoning in lightweight models through a two-stage debiasing process. First, the Bias Aware Training stage forges a dedicated "bias model" to embody these shortcut behaviors. Then, the Causal Debiasing Policy Optimization (CDPO) algorithm fine-tunes the primary model, employing an innovative repulsive objective to actively push it away from the bias model's flawed logic while simultaneously pulling it toward correct, generalizable solutions. Our model, VideoThinker-R1, establishes a new state-of-the-art in video reasoning efficiency. For same-scale comparison, requiring no Supervised Fine-Tuning (SFT) and using only 1% of the training data for RL, it surpasses VideoRFT-3B with a 3.2% average gain on widely-used benchmarks and a 7% lead on VideoMME. For cross-scale comparison, it outperforms the larger Video-UTR-7B model on multiple benchmarks, including a 2.1% gain on MVBench and a 3.8% gain on TempCompass. Code is available at https://github.com/falonss703/VideoThinker.