Explanation of deep learning models

Explainability should take into account both interpretability and completeness. The interpretability is the human understanding of the structure of a system. And the completeness is the ability to explain how a system works in an accurate way. In the explainable model based on deep learning, we will explain how to provide explanatory power in two aspects. Explainable models based on deep learning are roughly divided into two methods: Activation-Based Methods (ABMs) and Backpropagation-Based Methods (BBMs). Representative examples of ABMs include Class Activation Map and Gradient-Class Activation Map, and representative examples of BBMs include Layer-wise Relevance Propagation, Deep Learning Important Features, and SmoothGrad.