Improving Interpretability of CNN Models Using Non-Negative Concept Activation Vectors

Open in new window