ProtoPShare: Prototype Sharing for Interpretable Image Classification and Similarity Discovery
This work addresses the problem of interpretability in image classification for researchers and practitioners by improving the efficiency and robustness of prototype-based explanation methods.
This paper introduces ProtoPShare, a method for interpretable image classification that uses prototypical parts. The key innovation is an efficient data-dependent merge-pruning technique that allows for sharing prototypical parts between classes, leading to more consistent prototypes and improved robustness against image perturbations compared to ProtoPNet.
In this paper, we introduce ProtoPShare, a self-explained method that incorporates the paradigm of prototypical parts to explain its predictions. The main novelty of the ProtoPShare is its ability to efficiently share prototypical parts between the classes thanks to our data-dependent merge-pruning. Moreover, the prototypes are more consistent and the model is more robust to image perturbations than the state of the art method ProtoPNet. We verify our findings on two datasets, the CUB-200-2011 and the Stanford Cars.