Data processing system based on transverse federated learning

Abstract

The invention relates to a data processing system based on transverse federated learning, and the system comprises the steps: S1, initializing a global model, and setting K and a similarity threshold value Q; s2, selecting K clients, sending the initial global model for training, obtaining a first round of global model and a global update matrix, wherein t is set as 2; s3, selecting K clients, and sending a previous round of global model Gglobal, t-1 and a global update matrix [delta] Gglobal, t-1 to carry out tth round of training, and obtaining a t-th round of local model Gk, t and a local update matrix [delta] Gk, t of the kth client; s4, obtaining the similarity between the [delta] Gk, t and the [delta] Gglobal, t-1, and determining the client withe the similarity greater than Q as a target client; s5, re-matching the parameter sequence of the local model of the target client; s6, obtaining a t-th round of global model Gglobal, t and a global update matrix [delta] Gglobal, t; and S7, judging whether the global model is converged or not, if so, ending, otherwise, letting t = t + 1, and returning to S3. According to the system, the communication round number is reduced, the model convergence is accelerated, and the communication overhead is reduced.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to a data processing system based on horizontal federated learning. Background technique [0002] Federated learning is a machine learning framework. During model training, the data only exists on the local client of each participant. It does not gather all the data to a certain place for model training like traditional centralized learning. . Federated learning is divided into horizontal federated learning and vertical federated learning. Horizontal federated learning is also called federated learning by sample division and federated learning of feature alignment. In the existing cross-device horizontal federated learning model training process, network communication has become a major problem for technicians. Due to the instability of network signals of many terminal devices and the uneven network transmission capabilities of terminal devices, therefore, How to reduce the numb...

AI Technical Summary

Problems solved by technology

In the existing cross-device horizontal federated learning model training process, network communication has become a major problem for technicians. Due to the instability of network signals of many terminal devices and the uneven network transmission capabilities of terminal devices, therefore, How to reduce the number of communication rounds and speed up model convergence, thereby reducing communication overhead has become an urgent technical problem to be solved

Images