As a typical unsupervised learning model in neural networks, the self-encoder has attracted widespread attention in various areas, and its application in rolling bearing fault diagnosis is increasing with obvious advantages in data noise reduction and data visualization dimension reduction. In order to timely understand and master the application of auto-encoder and its improved algorithm in rolling bearing, this paper classifies and summarizes the representative auto-encoder related algorithms in recent years. Firstly, the principle of self-encoder and the theoretical sketch of several self-encoder methods based on its improvement are described, and the improvement purpose and improvement of these algorithms are analyzed. Then, the applications of these algorithms in the field of rolling bearing fault diagnosis are listed. Finally, the problems of present-day self-encoders and their improved algorithms are summarized, and the ideas for solving them are analyzed.