🤖 AI Summary
Existing image classification explanation methods produce only a single explanation per input, failing to comprehensively reveal the model’s decision logic and thereby limiting interpretability analysis and error diagnosis. To address this, we propose MultEX—the first multi-explanation generation framework for black-box classifiers—grounded in actual causality theory to formally model and control explanation diversity. MultEX integrates counterfactual perturbation analysis, saliency-constrained optimization, and efficient black-box query sampling. Evaluated on three mainstream model architectures and three standard benchmarks, MultEX outperforms state-of-the-art methods: it increases explanation count by 42%, while significantly improving explanation fidelity and human agreement. Crucially, MultEX establishes the first theoretically grounded, quantity-controllable, and semantically diverse multi-explanation paradigm for black-box image classification.
📝 Abstract
Existing explanation tools for image classifiers usually give only a single explanation for an image's classification. For many images, however, image classifiers accept more than one explanation for the image label. These explanations are useful for analyzing the decision process of the classifier and for detecting errors. Thus, restricting the number of explanations to just one severely limits insight into the behavior of the classifier. In this paper, we describe an algorithm and a tool, MultEX, for computing multiple explanations as the output of a black-box image classifier for a given image. Our algorithm uses a principled approach based on actual causality. We analyze its theoretical complexity and evaluate MultEX against the state-of-the-art across three different models and three different datasets. We find that MultEX finds more explanations and that these explanations are of higher quality.
