Research on Model Optimization Technology of Federated Learning

Federated learning is a new distributed machine learning model training method, which can protect user privacy while enabling a large number of edge devices to train a shared model based on their local data sets, and enable participants to benefit from each other. However, federated learning also faces some serious challenges, such as resource constraints, uneven data distribution, edge dynamics, and so on. Moreover, the existing federated learning methods may lead to long training time, consume a lot of bandwidth and other resources, and the heterogeneity of edge devices further worsens the inherent synchronization barrier problem of federated learning. In response to the above problems, this paper studies the existing model optimization technologies of federated learning, including node selection mechanism, local data enhancement, model compression, local update, gradient compression and other model optimization technologies, and summarizes and analyzes them.