CASPA：面向视觉语言模型模态无关适配的图结构概念锚点
CASPA: Graph-Structured Concept Anchors for Modality-Agnostic Adaptation in Vision-Language Models
把适配从「每类一组提示词参数」改为「共享文本/图像锚点上的软关联分布」，在冻结 CLIP 骨干的前提下实现参数共享与跨类语义复用；做 VLM 下游适配（少样本、跨数据迁移）时可直接替换 class-specific prompt 的参数化方式，并用 S-XCR 约束图文推理空间一致。
原文依据 we rethink adaptation as a shared conceptual structure rather than a per-class specialization
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论近期视觉语言模型（VLM）的进展同时揭示了大规模预训练的前景与刚性。尽管其零样本泛化能力令人印象深刻，现有适配范式——无论是提示词微调、适配器注入还是微调——仍然是类别特定的、模态偏置的且结构无关的。然而，这些设计选择限制了跨任务的推理级迁移。为此，我们将适配重新思考为一种共享的概念结构，而非逐类的专门化。我们提出 CASPA（概念锚定语义提示词适配器），这是一种双锚点语义适配器，联合学习共享的文本锚点与图像锚点，作为模态之间的双向概念接口。每个类别在这些锚点上学习一个软关联分布，产生组合式表示，从而实现参数共享与语义复用。为进一步对齐视觉与文本推理空间，CASPA 采用语义交叉一致性正则化（S-XCR），强制文本条件与图像条件的锚点混合之间在几何与语义上保持一致。因此，CASPA 在保持 CLIP 骨干网络冻结的同时，为类别条件提示词参数化提供了一种结构受约束的替代方案。在四种设定（基类到新类、跨数据迁移、少样本以及骨干无关评估）下，于十一个多样化的视觉识别数据集上进行评估，CASPA 达到或超越了最优方法。
英文原文摘要 1460 字符
Recent advances in vision-language models (VLMs) have revealed both the promise and the rigidity of large-scale pretraining. Despite their impressive zero-shot generalization, existing adaptation paradigms--whether prompt-tuning, adapter injection, or fine-tuning--remain class-specific, modality-biased, and structure-agnostic. However, these design choices limit reasoning-level transfer across tasks. To this end, we rethink adaptation as a shared conceptual structure rather than a per-class specialization. We propose CASPA (Concept-Anchored Semantic Prompt Adapter), a dual-anchor semantic adapter that jointly learns shared text and image anchors as a bidirectional conceptual interface between modalities. Each class learns a soft association distribution over these anchors, producing compositional representations which enable parameter sharing and semantic reuse. To further align visual and textual reasoning spaces, CASPA employs Semantic Cross-Consistency Regularization (S-XCR), enforcing geometric and semantic agreement between text- and image-conditioned anchor mixtures. CASPA, therefore, provides a structurally constrained alternative to class-conditional prompt parameterization while keeping the CLIP backbone frozen. Evaluated across four regimes, Base-to-Novel setup, cross-data transfer, few-shot, and backbone-agnostic evaluations, on eleven diverse visual recognition datasets, CASPA matches or outperforms state-of-the-art methods.