VAST：面向数据高效视频推理的视频能力分层分类体系
VAST: Video Ability-Stratified Taxonomy for Data-Efficient Video Reasoning
对做视频 MLLM 推理的人，这篇给出了按「感知-推理-认知」能力分层组织训练数据、并用一致性奖励做 RL 微调的配方，可在不改架构的前提下大幅降低 RL 训练所需的样本量与 GPU 时数，VAST-Bench 也可直接作为能力分层评测集使用。
原文依据 a reinforcement learning framework that uses consistency rewards to encourage alignment between reasoning traces and final answers, without architectural modifications
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论强化学习已成为提升多模态大语言模型视频推理能力的有效途径。然而，现有方法因两方面原因仍然效率低下。其一，训练数据通常按任务形式而非底层推理能力组织，这种错配促使模型学习任务特定的模式而非可迁移的能力。因此，提升能力泛化需要对大量能力-任务组合进行广泛覆盖，使强化学习训练代价高昂。其二，先前工作往往用日益复杂的算法设计来弥补这种低效，例如专门化的时序架构或多目标奖励框架，这进一步增加了训练复杂度。为解决这些问题，我们提出 VAST，一个认知分类体系与数据组织框架，将视频理解结构化为感知、推理、认知三个层次。基于该分类体系，我们构建了用于训练的 VAST-15K 与用于评测的 VAST-Bench。我们进一步提出 VideoVAST，一个使用一致性奖励来促进推理轨迹与最终答案之间对齐的强化学习框架，且无需对架构做任何修改。实验表明，在相同训练设置下，VideoVAST 在 MVBench 上达到 66.3% 的准确率、在 VAST-Bench 上达到 57.4%，而 Video-R1 分别为 62.7% 与 54.3%，同时 GPU 时数减少约 72%、训练样本减少 96%。代码与资源见 https://zhongan-wang.github.io/VAST。
英文原文摘要 1574 字符
Reinforcement learning (RL) has emerged as an effective approach for improving video reasoning in multimodal large language models (MLLMs). However, existing methods remain inefficient for two reasons. First, training data are typically organized by task formats rather than underlying reasoning abilities, creating a mismatch that encourages models to learn task-specific patterns instead of transferable capabilities. As a result, improving ability generalization requires broad coverage over many ability-task combinations, making RL training costly. Second, prior work often compensates for this inefficiency with increasingly complex algorithmic designs, such as specialized temporal architectures or multi-objective reward frameworks, which further complicate training. To address these issues, we present VAST, a cognitive taxonomy and data organization framework that structures video understanding into three layers: Perception, Reasoning, and Cognition. Based on this taxonomy, we construct VAST-15K for training and VAST-Bench for evaluation. We further introduce VideoVAST, a reinforcement learning framework that uses consistency rewards to encourage alignment between reasoning traces and final answers, without architectural modifications. Experiments show that VideoVAST achieves 66.3% accuracy on MVBench and 57.4% on VAST-Bench, compared with 62.7% and 54.3% for Video-R1, while using approximately 72% fewer GPU hours and 96% fewer training samples under the same training settings. Code and resources are available at https://zhongan-wang.github.io/VAST.