Interpretability of deep models

Class Activation Mapping (CAM)

Technique that generates discriminative localization maps for convolutional neural networks without supervision, using information from the final convolutional layer to spatially weight the importance of each input region for a given class.

← Indietro