Federal learning excitation method under specific indexes

Abstract

The invention provides a two-stage federated learning excitation method under a specific index. The method comprises the following steps: receiving a platform model precision improvement task index issued by a platform server; formulating a learning strategy according to a model precision improvement target issued by the platform server; training and acquiring the total reward amount of the platform server based on the learning strategy; and obtaining the reward amount allocated by the platform server based on the platform model precision value improvement contribution proportion. The two-stage federated learning incentive mechanism under the specific model precision index provided by the invention can be more combined with reality, unnecessary cost waste is reduced, and the incentive mechanism designed from the perspective of data quality and data quantity more comprehensively, scientifically and systematically improves the training efficiency of federated learning.

Description

technical field [0001] The invention provides a federated learning incentive method under a specific index, belongs to the field of distributed machine learning, and specifically provides a federated learning incentive method under a specific index. Background technique [0002] With the continuous development of machine learning technology, data security has become an inevitable problem, and federated learning as a new distributed machine learning model can well solve the problem of data privacy. The basic federated learning model solves the data privacy problem, but there is another problem with technologies like crowd sensing, that is, the collaboration between data islands and platform servers becomes inefficient. Therefore, it is common practice to design appropriate incentive mechanisms to maximize the benefits of each participant and society. [0003] The main research directions of federated learning incentive mechanism include Stackelberg game, auction, contract th...

AI Technical Summary

Problems solved by technology

That is, it is not combined with the actual situation, only for the purpose of obtaining the theoretical optimal solution, while ignoring the redundant problem of model accuracy in the actual operation process, which may lead to the problem of cost increase; the data quality and data quantity are not effectively used as incentive mechanisms based on

Images