Introduction of Explainable Machine Learning
Recently I start studying Introduction to Explainable Machine LearningExplainable Machine learning. There is a long history of interest in explainable machine learning. One can observe that most machine learning models, especially deep learning models, are usually black-box models. Specifically, with the deepening of the network layers and the obscurity of the mechanism, people can only observe the output of the machine learning model, but not understand the reason.
In this circumstance, it makes the machine learning model risky in the application process, such as automatic driving. Specifically, when the self-driving car slows down, one rarely knows the reason is road signs, pedestrians, or roads. In this way, it greatly reduces the credibility of the technology, thus affecting its further iterative upgrades. Therefore explainable machine learning catch more and more attention. The key questions for the black box of AI have been displayed in the below figure, which points out the credibility risks in the application process of AI. Then It broadens our horizons, and it overlaps with many applications, which may be an opportunity to do some research on it.
In summary, this blog introduces the need for explainable machine learning in short. Only when we can successfully master the mechanism of explainability for machine learning, can we explain it, understand it, improve it and trust it.