用于解释表征的对比语料归因方法
Contrastive Corpus Attribution for Explaining Representations
AI总结：
针对无监督模型表征缺乏语义难以解释的问题，提出基于参考语料与对比箔集的对比语料相似度，结合多种事后归因生成COCOA，经定量验证有效，可用于对比学习增强解释与CLIP零样本定位。
摘要：
尽管无监督模型已得到广泛应用，但专为这类模型设计的解释方法却极少。大多数解释方法用于解释标量形式的模型输出，然而无监督模型输出的是表征向量，其元素缺乏语义含义，并非合适的解释对象。为弥补这一缺口，近期研究定义了一种标量解释输出：表征空间中待解释样本（即被解释项）与其他样本基于点积的相似度。尽管这一方法实现了对无监督模型的解释，但其解读仍可能不够清晰，因为与被解释项表征的相似度对人类而言未必有实际意义。针对这一问题，我们提出对比语料相似度——一种基于参考语料和对比箔样本集的全新、具备语义意义的标量解释输出。我们证明对比语料相似度可与多种事后特征归因方法兼容，以生成对比语料归因（COntrastive COrpus Attributions，COCOA），并通过定量验证确认该方法能识别出对语料而言重要的特征。我们从两方面展示了COCOA的实用性：其一，在对比学习设置（SimCLR）中，通过解释同一图像的不同增强方式得出相关洞见；其二，通过解释图像表征与联合学习的文本表征（CLIP）的相似度，实现零样本目标定位。
英文摘要：
Despite the widespread use of unsupervised models, very few methods are designed to explain them. Most explanation methods explain a scalar model output. However, unsupervised models output representation vectors, the elements of which are not good candidates to explain because they lack semantic meaning. To bridge this gap, recent works defined a scalar explanation output: a dot product-based similarity in the representation space to the sample being explained (i.e., an explicand). Although this enabled explanations of unsupervised models, the interpretation of this approach can still be opaque because similarity to the explicand's representation may not be meaningful to humans. To address this, we propose contrastive corpus similarity, a novel and semantically meaningful scalar explanation output based on a reference corpus and a contrasting foil set of samples. We demonstrate that contrastive corpus similarity is compatible with many post-hoc feature attribution methods to generate COntrastive COrpus Attributions (COCOA) and quantitatively verify that features important to the corpus are identified. We showcase the utility of COCOA in two ways: (i) we draw insights by explaining augmentations of the same image in a contrastive learning setting (SimCLR); and (ii) we perform zero-shot object localization by explaining the similarity of image representations to jointly learned text representations (CLIP).
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
