Federated Learning Techniques for the Industrial Internet of Things
1Fuzhou University, Singapore
2Xidian University, Xi'an, China
3Georgia Southern University, Statesboro, USA
4Lulea University of Technology, Lulea, Sweden
Federated Learning Techniques for the Industrial Internet of Things
Description
The Industrial Internet of Things (IIoT) refers to the billions of physical devices (such as interconnected sensors, instruments, and other devices) around the world that are now connected to the internet without requiring human-to-human or human-to-computer interaction. It is advantageous in low deployment cost and extensive geographic coverage and has found numerous applications in diverse domains, including transportation, environment monitoring, smart city, pervasive healthcare, etc. Through low-cost computing, cloud, big data, analytics, and mobile technologies, IIoT devices can share and collect data with minimal human intervention. The data collected by IIoT devices can be transmitted to the central cloud server for storage. The security and privacy issues can be easily solved by data encryption; however, IIoT-based systems often face security and privacy issues, especially during data processing and analytic. Federated learning (FL) is considered a new computing paradigm that aims to train a machine learning algorithm on multiple local datasets in local mobile nodes without explicitly exchanging data samples. The general principle consists of training local models on local data samples and exchanging parameters (e.g., the weights and biases of a deep neural network) between these local nodes at some frequency to generate a global model shared by all mobile nodes.
However, federated learning solutions still face many new security and privacy challenges when it needs to meet the mobile IIoT paradigm. In the traditional FL model, the centric server aggregate submodel is locally trained by the distributed mobile node. The centric trained FL model will distribute to all mobile nodes in the system for usage. Moreover, the FL framework participators can start a membership inference attack to judge whether some specific participator is involved in the system for model training. In addition, the participator can launch an attack to exact the other participator's historical training data from the centric model. Furthermore, some malicious participators can upload a fake sub-model to damage the centric model, making the FL model fail.
The aim of this Special Issue is to solicit original research articles and review articles highlighting recent advanced security and privacy techniques relevant to the convergence of federated learning in IIoT-based systems. We hope that this Special Issue also gathers submissions discussing arising new challenges and opportunities from traditional federated learning architectures.
Potential topics include but are not limited to the following:
- Formal security model design in IIoT of data collection
- Lightweight secure data processing scheme for smart federated learning
- Software/hardware security for mobile IIoT devices
- Secure data validation techniques for distributing federated learning
- Trust IIoT device issues for federated learning
- Secure communication for IIoT-based federated learning
- Future and smart FL-based vulnerability assessment interfaces
- Secure programming and toolkits for federated learning
- Virtualisation and management for IIoT-based federated learning
- New secure computation framework for federated learning
- Secure distributed data storage for IIoT-based system
- Blockchain and smart contracts for IIoT-based federated learning
- New privacy challenges in IIoT-based federated learning