【Author】
Majeed, Umer; Hong, Choong Seon
【Source】2019 20TH ASIA-PACIFIC NETWORK OPERATIONS AND MANAGEMENT SYMPOSIUM (APNOMS)
【Abstract】In this paper, we propose blockchain network based architecture called FLchain for enhancing security of Federated Learning (FL). We leverage the concept of channels for learning multiple global models on FLchain. Local model parameters for each global iteration are stored as a block on the channel-specific ledger. We introduce the notion of the global model state trie which is stored and updated on the blockchain network based on the aggregation of local model updates collected from mobile devices. Qualitative evaluation shows that FLchain is more robust than traditional FL schemes as it ensures provenance and maintains auditable aspects of FL model in an immutable manner.
【Keywords】Blockchain; distributed computing; federated learning; multi-access edge computing
【标题】FLchain：通过支持 MEC 的区块链网络进行联邦学习
【摘要】在本文中，我们提出了基于区块链网络的架构，称为 FLchain，用于增强联邦学习 (FL) 的安全性。我们利用通道的概念来学习 FLchain 上的多个全局模型。每次全局迭代的局部模型参数都作为一个块存储在特定于通道的分类帐上。我们引入了全局模型状态树的概念，它基于从移动设备收集的本地模型更新的聚合，在区块链网络上存储和更新。定性评估表明，FLchain 比传统的 FL 方案更稳健，因为它确保了来源并以不可变的方式维护了 FL 模型的可审计方面。
【关键词】区块链；分布式计算；联邦学习；多接入边缘计算
评论