Excitation method and system based on reverse auction

Abstract

The invention discloses an incentive method and system based on reverse auction. The method comprises the steps that a data user submits task information to a platform; the platform audits the task submitted by the data user and then issues the task to the potential participants; the participants bid the published tasks; the platform selects the lowest price and the bidder of each task as the price and the winner of the corresponding task until no new bid exists; the task with the lowest bid of each task obtains the task execution right; the platform pre-pays a part of reward to the participant who obtains the task execution right, the participant executes the task and uploads data to the platform after completing the task, the platform settles the reward to the participant, and the data user pays the reward to the platform. According to the excitation method and system based on reverse auction, the problem that a curse is won is solved, and compared with similar methods, the method has higher auction efficiency, participant utility, platform utility and social utility, and the satisfaction degree of all subjects in federated learning is improved.

Description

technical field [0001] The invention relates to the technical field of federated learning, and in particular provides an incentive method and system based on a reverse auction. Background technique [0002] Federated learning can efficiently use mobile phones to collect data and provide services, and its feasibility and incentive mechanism have been extensively studied. Combining reverse auction and Vickrey auction, Zhu et al proposed the RVA auction incentive method, using the auction method to ensure the fairness of task allocation, and effectively realize the task allocation problem of federated learning. However, we found that the cost assumptions in previous studies, including RVA, are not realistic, leading to the phenomenon that the cost of the participants is less than the reward and loses money in the actual application of the theoretically reasonable incentive mechanism, that is, the winner's curse. This phenomenon makes the winners in the auction lose money after...

AI Technical Summary

Problems solved by technology

However, we found that the cost assumptions in previous studies, including RVA, are not realistic, leading to the fact that the theoretically reasonable incentive mechanism will cause the participant’s cost to be less than the reward and lose money in practice, that is, the winner’s curse

This phenomenon makes the winners in the auction lose money after completing the federated learning tasks, which leads to negative incentives and is not conducive to the development of federated learning

In addition, it is found that most studies including RVA assume that a participant can only participate in one task, ignoring the development of networks and sensors to enable the rapid development of federated learning, and the future development trend that a single participant can participate in multiple perception tasks at the same time

Images