让音频语言模型在少量样本下更好泛化，避免过拟合已有类别。
SubT: Subspace Tuning for Few-shot Generalization of Audio-Language Models
- 通过嵌入空间几何约束与零样本原型锚定，保持类别结构
- 11个音频基准上实现强少样本泛化能力，无需反向传播文本编码器
- 适合需要高效适配新类别的音频模型应用
预训练音频-语言模型（ALMs）的少样本参数高效适配常提升已见类性能，却损害未见类泛化能力，形成基线到新类的权衡。我们通过文本嵌入空间中的零样本漂移分析该问题：少样本微调会破坏类别间结构，使适配后的嵌入偏离预训练锚点。为此提出子空间调优（SubT），一种结合几何感知共享变换与零样本原型锚定的嵌入空间适配方法。学习到的变换可迁移至未见类别，通过子空间感知门控缓解负迁移。在11个音频基准上，SubT在直接操作预计算文本嵌入的前提下，实现强少样本泛化能力，无需文本编码器反向传播。代码开源于https://github.com/jhyukjang/SubT。
原文摘要 · Abstract (English)
Few-shot parameter-efficient adaptation of pretrained Audio--Language Models (ALMs) often improves seen-class performance at the cost of unseen-class generalization, leading to the base-to-new trade-off. We study this failure through zero-shot drift in the text embedding space: few-shot tuning can distort inter-class structure and move adapted embeddings away from their pretrained anchors. We therefore propose Subspace Tuning (SubT), an embedding-space adaptation method that combines a geometry-aware shared transformation with anchoring to the zero-shot prototypes. The learned transformation is transferred to unseen classes, with subspace-aware gating to mitigate negative transfer. Across 11 audio benchmarks, SubT achieves strong few-shot generalization while operating directly on precomputed text embeddings without text-encoder backpropagation. Code is available at https://github.com/jhyukjang/SubT.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。