Optimization of Federated Learning Communications with Heterogeneous Quantization
Peng Liu, Tianli Gao, Congduan Li · 2022
The rapid development of machine learning in the field of artificial intelligence benefits from a large amount of training data. Due to the problems of data fragmentation and data isolation, federated learning was proposed. However, there is a communication bottleneck in the learning process of federated learning. To solve this problem, this paper adopts the method of quantization to optimize the communication of federated learning and quantifies features with different accuracy according to the feature importance. We compare the communication overhead between the communication optimization method and the non-communication optimization method, and give a theoretical explanation based on the scenario of detecting fraud in bank credit card transactions.