Switched linear projections and inactive state sensitivity for deep neural network interpretability

Open in new window