A Efficient Prototype-Assisted Clustered Federated Learning Framework for Industrial Internet of Things
Fuyao Zhang, Dan Wang, Yalan Jiang · 2023
The Industrial Internet of Things (IIoT) is a critical enabler of the industrial future and offers a novel paradigm for Industry 4.0 concepts. Federated learning (FL) technology is applied to the IIoT as a distributed machine learning method. It can protect user privacy by harmonising multiple IIoT equipment for AI model training at the edge of the network, while ensuring that no locally sensitive data is exchanged. However, FL is still challenged by non-independent identical distribution (non-IID). Due to the uncertainty of information distribution, model fusion may have parameter mismatches, resulting in model structure misalignment and many other problems. To address the above issues, we propose an Efficient Prototype-assisted Clustered Federated Learning Framework (PClusterFL) for IIoT in the field of cluster federation, which adopts the cosine distance of the prototype as a similarity measure instead of the distance between model parameters, thereby reducing the communication burden. In addition, we design a prototype-assisted aggregation strategy to improve aggregation accuracy. Numerous experiments are performed between PClusterFL and the latest baseline method. Experimental results indicate that PClusterFL is better than several state-of-the-art FL methods on several datasets.