ClusterFL: A Similarity-Aware Federated Learning System for Human Activity Recognition

Federated Learning (FL) has recently received significant interests thanks to its capability of preserving data privacy. However, existing FL paradigms yield unsatisfactory performance for a wide class of human activity recognition (HAR) applications since they are oblivious to the intrinsic relationship between data of different users.

We propose ClusterFL, a similarity-aware federated learning system that can provide high model accuracy and low communication overhead for HAR applications. ClusterFL features a novel clustered multi-task federated learning framework that maximizes the training accuracy of multiple learned models while automatically capturing the intrinsic clustering relationship among the data of different nodes. Based on the learned cluster relationship, ClusterFL can efficiently drop out the nodes that converge slower or have little correlation with other nodes in a cluster, significantly speeding up the convergence while maintaining the accuracy performance.

We evaluate the performance of ClusterFL on a NVIDIA edge testbed using four new HAR datasets collected from total 145 users. The results show that, ClusterFL outperforms several state-of-the- art FL paradigms in terms of overall accuracy, and save more than 50% communication overhead at the expense of negligible accuracy degradation.

Funding

1. Hong Kong General Research Fund, “HomeSense: A Pervasive System for Home Activity Recognition via Federated Learning”, PI, HKD $810K, 2021/01/01-2023/12/31.
2. Alzheimer’s Drug Discovery Foundation(ADDF), “Machine Learning Technologies for Advanced Digital Biomarkers for Alzheimer's Disease”, PI, HKD $5.6 million, 2021-2023.

People

Guoliang Xing (PI, Professor, CUHK)
Xiaomin OUYANG (Ph.D candidate, CUHK)
Zhiyuan XIE (Ph.D candidate, CUHK)

Collaborator

Jianwei HUANG (Professor, CUHK-SZ)
Jiayu ZHOU (Associate Professor, MSU)