无需训练，用扩散模型快速生成特征不变集样例，验证模型是否该变不变。
Show Me What You Don't Know: Efficient Sampling from Invariant Sets for Model Validation
- 用纤维损失引导扩散过程，无须训练即可生成特征不变集样本。
- 在ImageNet和CheXpert数据集上验证，发现模型对解剖异常的不敏感现象。
- 适合评估医学影像模型的鲁棒性，尤其关注关键病理特征的敏感性。
机器学习模型的性能取决于其学习特征的质量：应忽略无关数据变化而对任务相关细节敏感。为可视化这种特性，我们提出一种方法，通过从特征提取器的纤维（即由不变性定义的等价类）中采样来分析其行为。不同于以往需为每个特征检测器单独训练生成模型的工作，我们的算法无需训练，直接利用预训练的扩散模型或流匹配模型作为先验。通过纤维损失（惩罚特征不匹配）引导去噪过程，实现非线性扩散轨迹匹配，将原本需数天训练的不变性学习替换为一次高效生成。在ImageNet、CheXpert等主流数据集及ResNet、DINO、BiomedClip等模型上的实验表明，该框架可揭示从理想到令人担忧的各种不变性行为。例如，我们发现Qwen-2B将倒置心位（situs inversus）患者与正常解剖结构置于同一纤维中。
原文摘要 · Abstract (English)
The performance of machine learning models is determined by the quality of their learned features. They should be invariant under irrelevant data variation but sensitive to task-relevant details. To visualize whether this is the case, we propose a method to analyze feature extractors by sampling from their fibers -- equivalence classes defined by their invariances -- given an arbitrary representative. Unlike existing work where a dedicated generative model is trained for each feature detector, our algorithm is training-free and exploits a pretrained diffusion or flow-matching model as a prior. The fiber loss -- which penalizes mismatch in features -- guides the denoising process toward the desired equivalence class, via non-linear diffusion trajectory matching. This replaces days of training for invariance learning with a single guided generation procedure at comparable fidelity. Experiments on popular datasets (ImageNet, CheXpert) and model types (ResNet, DINO, BiomedClip) demonstrate that our framework can reveal invariances ranging from very desirable to concerning behaviour. For instance, we show how Qwen-2B places patients with situs inversus (heart on the right side) in the same fiber as typical anatomy.
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。