保护以适配：面向少样本动作识别的正交子空间控制与排序负提示词课程
Protect to Adapt: Orthogonal Subspace Control with Ranked Negative-Prompt Curriculum for Few-Shot Action Recognition
OSC 把低秩更新约束在预训练骨干网络主语义子空间的正交补上，这一做法可直接迁移到任何 VLM 微调场景（如少样本图像分类、视频检索），在适配下游任务的同时抑制灾难性遗忘；RNC 用大语言模型生成按难度排序的负提示词，为少样本 episode 中负样本不足、决策边界模糊的问题提供了一种可复用的数据构造思路。
原文依据 constrains low-rank updates to its orthogonal complement, preserving domain-general semantics while allowing task-specific adaptation
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论将视觉语言模型（VLM）适配到少样本动作识别（FSAR）往往以牺牲稳定性换取精度：任务特定的收益可能引发对领域通用知识的灾难性遗忘，并缩小类间间隔。在少样本 episode 中，每个查询仅与一个正类以及少量负类进行对比，因此文本编码器看到的提示词多样性有限，且很少观察到决策边界附近的困难反例。我们提出 Protect-to-Adapt（P2A），一种参数高效的微调方法，包含两个互补模块。正交子空间控制（OSC）估计预训练骨干网络的主语义子空间，并将低秩更新约束到其正交补上，从而在允许任务特定适配的同时保留领域通用语义。排序负提示词课程（RNC）利用大语言模型生成经校验器过滤、难度递增的负提示词。这些类特定的困难反例在少样本条件下扩大了间隔并使决策边界更加锐利。在仅训练骨干网络 2% 参数的情况下，P2A 在五个 FSAR 基准上取得最优性能，并在跨数据集的持续学习设定中大幅减少灾难性遗忘，该设定下模型被顺序适配到多个视频数据集且不使用回放。
英文原文摘要 1346 字符
Adapting Vision-Language Models (VLMs) to few-shot action recognition (FSAR) often trades accuracy for stability: task-specific gains can trigger catastrophic forgetting of domain-general knowledge and reduce inter-class margins. In few-shot episodes, each query is contrasted with only one positive class and a few negatives, so the text encoder sees limited prompt diversity and rarely observes hard counter-examples near decision boundaries. We propose Protect-to-Adapt (P2A), a parameter-efficient fine-tuning method with two complementary modules. Orthogonal Subspace Control (OSC) estimates a principal semantic subspace of the pre-trained backbone and constrains low-rank updates to its orthogonal complement, preserving domain-general semantics while allowing task-specific adaptation. Ranked Negative-prompt Curriculum (RNC) uses a large language model to generate verifier-filtered negative prompts with increasing difficulty. These class-specific hard counter-examples enlarge margins and sharpen decision boundaries under few-shot conditions. With only 2% of backbone parameters trainable, P2A achieves state-of-the-art performance on five FSAR benchmarks and substantially reduces catastrophic forgetting in a cross-dataset continual-learning setting where the model is adapted sequentially to multiple video datasets without replay.