让AI说出判断图像的关键依据，解释更准确。
Zero-Shot Textual Explanations via Translating Decision-Critical Features
- 先找影响分类的神经元，再聚焦这些关键特征
- 生成的解释与模型真实决策逻辑匹配度更高
- 适合需要可信AI解释的医疗、金融等场景
文本解释通过自然语言描述图像分类器的预测理由，提升决策透明度。大型视觉语言模型虽能生成描述，但设计用于通用视觉理解，而非分类器特定推理。现有零样本解释方法对齐全局图像特征与语言，生成的是可见内容描述，而非驱动预测的关键因素。本文提出TEXTER，通过在对齐前识别决策关键特征来克服此局限。TEXTER定位对预测有贡献的神经元，并强调这些神经元编码的特征——即决策关键特征。随后将这些强化特征映射至CLIP特征空间，检索反映模型推理过程的文本解释。稀疏自编码器进一步提升了Transformer架构下的可解释性。大量实验表明，TEXTER生成的解释比现有方法更忠实、更易懂。代码已公开于https://github.com/tttt-0814/TEXTER。
原文摘要 · Abstract (English)
Textual explanations make image classifier decisions transparent by describing the prediction rationale in natural language. Large vision-language models can generate captions but are designed for general visual understanding, not classifier-specific reasoning. Existing zero-shot explanation methods align global image features with language, producing descriptions of what is visible rather than what drives the prediction. We propose TEXTER, which overcomes this limitation by isolating decision-critical features before alignment. TEXTER identifies the neurons contributing to the prediction and emphasizes the features encoded in those neurons -- i.e., the decision-critical features. It then maps these emphasized features into the CLIP feature space to retrieve textual explanations that reflect the model's reasoning. A sparse autoencoder further improves interpretability, particularly for Transformer architectures. Extensive experiments show that TEXTER provides more faithful and interpretable explanations than existing methods. The code is available at \url{https://github.com/tttt-0814/TEXTER}.
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。