In recent years, deep learning proves extremely effective at learning nonlinear features and functions from complex data, and has been widely used in text recognition [1], social networking [2], biomedical [3] and other fields. The effect of the deep learning model is often related to the size of the model and the training data set. Under a reasonable learning mechanism, if there is more training data, the model will have a better effect. However, in the era of big data, data is often scattered among individuals and cannot be brought together because of policy, competition, or privacy. Users can only learn based on only a part of data and cannot benefit from the whole data. In order to solve this problem, it is the current trend to apply collaborative learning to the deep learning. Collaborative deep learning is a situation in which two or more users learn a deep learning model together. The generation of collaborative deep learning avoids the problem of the long acquisition cycle of traditional deep learning data and the low accuracy of the model caused by the use of only a portion of the data. Although collaborative deep learning applications are becoming more and more widely used, due to the variety of collaborative deep learning application scenarios, privacy exposure methods are more diversified. For example, smart bracelets can record information such as the user’s heart rate and motion trajectory throughout the day. Smart homes can record the user’s diet, routines and other laws. These data can be collected to provide users with high-quality personalized services such as recommendation and identification. However, they also face unavoidable problems such as non-trusted third parties and untrusted users. How to ensure the utility of collaborative deep learning without revealing the privacy of users and models is a key issue in the field of deep learning. In this paper, we first introduce the architecture of collaborative deep learning and the issue of privacy leakage. Secondly, we introduce privacypreserving technology commonly used in the applications and analyze their advantages and disadvantages when using in the two phases of collaborative deep learning. Finally, we summarize its development direction and trend.