CHiQPM: Calibrated Hierarchical Interpretable Image Classification

TL;DR

This work proposes the Calibrated Hierarchical QPM (CHiQPM) which offers uniquely comprehensive global and local interpretability, paving the way for human-AI complementarity.

Abstract

Globally interpretable models are a promising approach for trustworthy AI in safety-critical domains. Alongside global explanations, detailed local explanations are a crucial complement to effectively support human experts during inference. This work proposes the Calibrated Hierarchical QPM (CHiQPM) which offers uniquely comprehensive global and local interpretability, paving the way for human-AI complementarity. CHiQPM achieves superior global interpretability by contrastively explaining the majority of classes and offers novel hierarchical explanations that are more similar to how humans reason and can be traversed to offer a built-in interpretable Conformal prediction (CP) method. Our comprehensive evaluation shows that CHiQPM achieves state-of-the-art accuracy as a point predictor, maintaining 99% accuracy of non-interpretable models. This demonstrates a substantial improvement, where interpretability is incorporated without sacrificing overall accuracy. Furthermore, its calibrated set prediction is competitively efficient to other CP methods, while providing interpretable predictions of coherent sets along its hierarchical explanation.

Figures (34)

• Figure 1: Contrastive global Explanation, comparing the class representations of Shiny and Bronzed Cowbirds for CHiQPM that represents every class with 3 of 30 features. The cowbirds are differentiated based on the red eye.
• Figure 2: Exemplary local explanation provided by our CHiQPM, with the global explanation in \ref{['fig:TeaserGlob']}, for a difficult test image of a Bronzed Cowbird with a pale red eye that is not clearly visible. This leads to negligible activation of the red-eye detecting Feature 7. The calibrated CHiQPM provides a hierarchical explanation that communicates clear evidence for the predicted coherent set of black birds (marked in bold with green edges, including the correct label), but no sufficient evidence to differentiate between them.
• Figure 3: Coverage relative to the set size on for various set prediction methods applied to CHiQPM with $5$ out of a total of $50$ assigned features per class. The stars denote different calibration or hierarchy levels and are linearly interpolated. Traversing the hierarchical explanations (\ref{['fig:TeaserFull']}), the built-in conformal prediction method predicts coherent sets with competitive efficiency to CP methods THR thrpaper or APS apspaper.
• Figure 4: Overview of our proposed pipeline to obtain a CHiQPM
• Figure 5: Gradient on features $\boldsymbol{f}^*$ for a train sample labeled GT for a toy example with 3 classes and 7 features, with $\boldsymbol{W}{}^*$ shown left. At the average activation on the CUB dataset, the Ground Truth Exclusive (GTE) feature has a roughly $4000$ times higher gradient than the other assigned features, which are shared with Sim, Ground Truth Shared (GTS).