Generating Part-Based Global Explanations Via Correspondence
This addresses the need for cost-effective global explanations in AI interpretability, though it is incremental as it builds on existing part-based and concept-based methods.
The paper tackles the problem of generating global explanations for deep learning models by leveraging user-defined part labels from a limited set of images to efficiently transfer them to a larger dataset, enabling scalable human-understandable explanations.
Deep learning models are notoriously opaque. Existing explanation methods often focus on localized visual explanations for individual images. Concept-based explanations, while offering global insights, require extensive annotations, incurring significant labeling cost. We propose an approach that leverages user-defined part labels from a limited set of images and efficiently transfers them to a larger dataset. This enables the generation of global symbolic explanations by aggregating part-based local explanations, ultimately providing human-understandable explanations for model decisions on a large scale.