CRAFT (Concept Recursive Activation FacTorization for Explainability) is a tool developed by a team of researchers from the Carney Institute for Brain Science at Brown University and some computer scientists from the Artificial and Natural Intelligence Toulouse Institute, France to demystify deep neural networks and uncover the strategies employed by AI systems to process images. CRAFT is an attribution method which generates heatmaps indicating the significance of individual pixels in influencing a model’s decision, however, it has certain limitations as it tends to focus solely on the most salient regions of an image. The tool aims to discern the “what” and “where” an AI model focuses on during the decision-making process, thereby providing a better understanding of the inner workings of such systems.
Related Posts
