In this article, we propose a trustworthy privacy-preserving federated learning (FL)-based deep learning (DL) service framework for Industrial Internet of Things-enabled systems. FL mitigates the privacy issues of the traditional collaborative learning model by aggregating multiple locally trained models without sharing any datasets among the participants. Nevertheless, the FL-based DL (FDL) model cannot be trusted as it is susceptible to intermediate results and data structure leakage during the model aggregation process. The proposed framework introduces an edge and cloud-powered service-oriented architecture identifying the key components and a service model for residual networks-based FDL with differential privacy for generating trustworthy locally trained models. The service model decomposes the functionality of the overall FDL process as services to ensure trustworthy execution through privacy preservation. Finally, we develop a privacy-preserving local model aggregation mechanism for FDL. We perform several experiments to assess the performance of the proposed framework.

中文翻译：

---

面向工业物联网的可信赖和隐私保护联邦深度学习服务框架

在本文中，我们为支持工业物联网的系统提出了一个基于可信隐私保护联邦学习 (FL) 的深度学习 (DL) 服务框架。FL 通过聚合多个本地训练的模型而不在参与者之间共享任何数据集来缓解传统协作学习模型的隐私问题。然而，基于 FL 的 DL (FDL) 模型不可信，因为它在模型聚合过程中容易受到中间结果和数据结构泄漏的影响。所提出的框架引入了一种边缘和云驱动的面向服务的架构，用于识别关键组件和基于残差网络的具有差异隐私的 FDL 的服务模型，用于生成可信赖的本地训练模型。服务模型将整个 FDL 流程的功能分解为服务，以通过隐私保护确保可信执行。最后，我们为 FDL 开发了一种保护隐私的本地模型聚合机制。我们进行了几个实验来评估所提出框架的性能。