Explaining Image Classifiers by Counterfactual Generation
This provides more interpretable explanations for image classifiers, which is important for users needing to understand model decisions, though it is an incremental improvement over existing saliency methods.
The paper tackles the problem of explaining image classifier predictions by identifying which image regions would most change the decision if removed, using a generative model to sample plausible in-fills instead of ad-hoc methods like blurring. The result is more compact and relevant saliency maps with fewer artifacts compared to previous approaches.
When an image classifier makes a prediction, which parts of the image are relevant and why? We can rephrase this question to ask: which parts of the image, if they were not seen by the classifier, would most change its decision? Producing an answer requires marginalizing over images that could have been seen but weren't. We can sample plausible image in-fills by conditioning a generative model on the rest of the image. We then optimize to find the image regions that most change the classifier's decision after in-fill. Our approach contrasts with ad-hoc in-filling approaches, such as blurring or injecting noise, which generate inputs far from the data distribution, and ignore informative relationships between different parts of the image. Our method produces more compact and relevant saliency maps, with fewer artifacts compared to previous methods.