An Efficient Personalized Federated Learning System and Method

Abstract

The present invention relates to an efficient personalized federated learning system and method. The terminal equipment in the system includes: a terminal equipment data module, a terminal equipment model download module, two terminal equipment model training modules and a terminal equipment model upload module. The central server includes A server-side data module, two server-side model integration modules and a server-side model distribution module. The invention has a reasonable design, and it puts the pruning process and model training in the terminal equipment completely, reduces the burden on the central server, improves the processing efficiency, and fully considers the difference of data distribution, realizes the personalized function of the model, and can Effectively analyze the data collected on different terminal devices, while greatly reducing communication costs while ensuring user privacy and personalization, as well as sending new models when terminal device data is missing.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence, and relates to a federated learning system, in particular to an efficient and personalized federated learning system and method. Background technique [0002] The success of AI relies heavily on large amounts of high-quality data. For example, the analysis and evaluation of customer behavior often relies on the combined analysis of a large number of heterogeneous data. This kind of data not only has a large amount of data, but also has extremely high personalization and privacy information. The regulatory environment at home and abroad is gradually strengthening data protection, so the free flow of data under the premise of security and compliance has become a general trend. In addition, from the perspective of users and enterprises, the data owned by commercial companies often have huge potential value. Based on interests, these institutions will not provide their own data to ...

AI Technical Summary

Problems solved by technology

[0004] However, the current development of federated learning has encountered two bottlenecks: the first is communication overhead, because a large number of terminal devices need to communicate with the server periodically, which naturally brings huge communication costs; the second is heterogeneous data, The data types are different, and the amount of data is also different. Due to the significant differences in the data collected by each terminal device, such as quantity, data category, data characteristics, etc., the data distribution between devices no longer meets the conditions of independent and identical distribution.

The central server cannot take care of the data on each terminal device, which significantly reduces the accuracy of the machine learning model obtained through traditional federated learning

And because traditional federated learning is mostly a global generalization model on all terminal devices, when one device or the central server is cracked, it will cause privacy and security risks to the data of other devices

[0005] To sum up, existing federated learning technologies generally have problems such as high communication costs, low degree of personalization, poor privacy protection, and reduced prediction accuracy due to insufficient or missing data.

Images