explainability

Concept Bottleneck Models (CBMs) have recently been proposed to address the ‘black-box’ problem of deep neural networks, by first mapping images to a human-understandable concept space and then linearly combining concepts for classification. Such …