Diffexplainer: Towards Cross-modal Global Explanations with Diffusion Models
This work addresses the need for cross-modal, human-interpretable explanations in AI, particularly for identifying biases and spurious features without manual intervention, though it is incremental in leveraging existing language-vision models.
The paper tackles the problem of providing global explanations for classifier decisions by introducing DiffExplainer, a framework that uses diffusion models conditioned on text prompts to synthesize images that maximize class outputs and hidden features, enabling automated bias identification and surpassing existing activation maximization methods.
We present DiffExplainer, a novel framework that, leveraging language-vision models, enables multimodal global explainability. DiffExplainer employs diffusion models conditioned on optimized text prompts, synthesizing images that maximize class outputs and hidden features of a classifier, thus providing a visual tool for explaining decisions. Moreover, the analysis of generated visual descriptions allows for automatic identification of biases and spurious features, as opposed to traditional methods that often rely on manual intervention. The cross-modal transferability of language-vision models also enables the possibility to describe decisions in a more human-interpretable way, i.e., through text. We conduct comprehensive experiments, which include an extensive user study, demonstrating the effectiveness of DiffExplainer on 1) the generation of high-quality images explaining model decisions, surpassing existing activation maximization methods, and 2) the automated identification of biases and spurious features.