面向视频-语言理解持续学习的可供性优先分解
Affordance-First Decomposition for Continual Learning in Video-Language Understanding
把视频映射到缓慢变化的 affordance token 作为共享时间对齐基底、只让轻量查询路由调度器做适配，这一「稳定基底 + 定向可塑」的拆分思路可直接迁移到视频-语言模型的持续学习与免回放增量训练中。
原文依据 videos are mapped to slowly varying affordance tokens that form a shared, time-aligned substrate
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论随着模型面对非平稳的数据、领域与查询风格，视频-语言理解的持续学习日益重要，然而主流方案模糊了哪些部分应保持稳定、哪些部分应进行适配，依赖静态路由/容量，或需要回放过去的视频。我们旨在现实的内存与隐私约束下，明确界定稳定性存在于何处、可塑性应聚焦于何处。我们提出可供性优先分解（AFD）：将视频映射为缓慢变化的可供性 token，构成共享的、时间对齐的基底，而一个轻量、查询路由、冲突感知的调度器集中进行适配，并仅在需要时扩展容量。该基底通过弱对齐与教师一致性得到稳定，训练采用仅问题回放。AFD 在各协议上均达到最优：在领域增量 VideoQA 上取得 51.6% 的平均准确率与 -1.8% 的遗忘率，ViLCo 上 R@1@0.5 为 29.6%（MQ）与 20.7%（NLQ），stAP@0.25 为 18.4%（VQ），在时间增量 iVQA 上取得 39.5% 的准确率与 -1.6% 的遗忘率。总体而言，AFD 在稳定的、以交互为中心的基底与定向适配之间提供了显式且可解释的划分。
英文原文摘要 1196 字符
Continual learning for video--language understanding is increasingly important as models face non-stationary data, domains, and query styles, yet prevailing solutions blur what should stay stable versus what should adapt, rely on static routing/capacity, or require replaying past videos. We aim to explicitly specify where stability lives and where plasticity should be focused under realistic memory and privacy constraints. We introduce Affordance-First Decomposition (AFD): videos are mapped to slowly varying affordance tokens that form a shared, time-aligned substrate, while a lightweight, query-routed, conflict-aware scheduler concentrates adaptation and grows capacity only when needed. The substrate is stabilized via weak alignment and teacher consistency, and training uses question-only replay. AFD achieves state-of-the-art across protocols: 51.6% average accuracy with -1.8% forgetting on domain-incremental VideoQA, ViLCo R@1@0.5 of 29.6% (MQ) and 20.7% (NLQ) with 18.4% stAP@0.25 (VQ), and 39.5% accuracy with -1.6% forgetting on time-incremental iVQA. Overall, AFD offers an explicit, interpretable split between a stable interaction-centered substrate and targeted adaptation.