Using prototypes to improve convolutional networks interpretability

Thalita F Drumond (1, 2, 3) , Thierry Viéville (1, 2, 3) , Frédéric Alexandre (1, 2, 3)

1 Mnemosyne - Mnemonic Synergy
2 IMN - Institut des Maladies Neurodégénératives [Bordeaux]
3 LaBRI - Laboratoire Bordelais de Recherche en Informatique

We propose a method that allows the interpretation of the data representation obtained by CNN, through introducing prototypes in the feature space, that are later classified into a certain category. This way we can see how the feature space is structured in link with the categories and the related task.

Domaines: Neurosciences

Thalita F Drumond, Thierry Viéville, Frédéric Alexandre. Using prototypes to improve convolutional networks interpretability. NIPS 2017 - 31st Annual Conference on Neural Information Processing Systems: Transparent and interpretable machine learning in safety critical environments Workshop, Dec 2017, Long Beach, United States. ⟨hal-01651964⟩