DFedSN: Decentralized federated learning based on heterogeneous data in social networks

Users talk with each other and share their lives online, which forms a huge social network. However, a series of potential problems such as privacy security and data leakage will occur in the process of data transmission and sharing. Therefore, we came up with the idea of putting social networks into the framework of decentralized federated learning. Federated learning is a model of distributed training model, which can limit users' own data in and out of the local area during transmission. Moreover, the decentralized federated learning eliminates the hidden threat from third-party servers. However, the data generated by a large user group is also messy and heterogeneous, and there are different distribution changes among them, which will greatly reduce the learning performance of the model. This paper aims to solve the problem of user data heterogeneity in decentralized federated learning, so as to reduce the loss of model performance. To be detailed, we set up the affine distribution of user data structure to capture the users who have the heterogeneity between data, and then put forward an approach called DFedSN training to reduce the independent identically distributed data loss. This method has excellent robustness even in the case of different data distribution, and it can train and learn the model stably without losing efficiency. Through several experiments on DFedSN algorithm in different neural networks, the results show that our algorithm can still achieve excellent training results even when there is a huge gap in the distribution of user data, and it will have no negative effects due to different distributions, which is the first step in decentralized federated learning.