A decentralized federated learning architecture is proposed to apply to the Businesses-to-Businesses scenarios by introducing the consortium blockchain in this paper. We introduce a model verification mechanism to ensure the quality of local models trained by participators. To analyze the latency of the system, a latency model is constructed by considering the work flow of the architecture. Finally the experiment results show that our latency model does well in quantifying the actual delays.
翻译:通过在本文件中引入财团链块,提议建立一个分散化的联邦学习结构,以适用于企业对企业的设想。我们引入了一种示范核查机制,以确保参加方培训的当地模型的质量。为了分析该系统的潜伏度,通过考虑结构的工作流程来构建一个潜伏模型。最后,实验结果表明,我们的潜伏模型在量化实际延误方面效果良好。