Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)
The interpretation of deep learning models is a challenge due to their size, complexity, and often opaque internal state. In addition, many systems, such as image classifiers, o...