Workflow from Scientific Research

Multi-class counterfactual explanation of dimension V =2. Left: query image of a ‘Scarlet Tanager’ (upper left) and two images randomly selected from the counterfactual class set C = {‘Red Headed Woodpecker,’ ‘Summer Tanager’}. Middle-left: attributive maps are computed for the query and each of counter images, with respect to class predictions hy* and hyv, yv in C and confidence predictor s. Middle-right: attributive maps are combined with (14) to generate discriminant explanations. Right: discriminant explanations are thresholded to generate a multi-class counterfactual explanation.

DOI