A Review of Explainable Artificial Intelligence

Artificial intelligence developed rapidly, while people are increasingly concerned about internal structure in machine learning models. Starting from the definition of interpretability and historical process of interpretability model, this paper summarizes and analyzes the existing interpretability methods according to the two dimensions of model type and model time based on the objectives of interpretability model and different categories. With the help of the existing interpretable methods, this paper summarizes and analyzes its application value to the society analyzes the reasons why its application is hindered. This paper concretely analyzes and summarizes the applications in industrial fields, including model debugging, feature engineering and data collection. This paper aims to summarizes the shortcomings of the existing interpretability model, and proposes some suggestions based on them. Starting from the nature of interpretability model, this paper analyzes and summarizes the disadvantages of the existing model evaluation index, and puts forward the quantitative evaluation index of the model from the definition of interpretability. Finally, this paper summarizes the above and looks forward to the development direction of interpretability models.