FedDSHAR: A dual-strategy federated learning approach for human activity recognition amid noise label user

Ziqian Lin, Xuefeng Jiang, Kun Zhang, Chongjun Fan, Yaya Liu

Abstract: Highlights•We present a new robust FL framework aimed at addressing complex label noise issue within real-world HAR datasets.•We introduce the innovative dual-strategy training scheme to enhance the exploitation of noisy data from users in the federated learning scenario.•We propose a novel strategy based on the concept of enriching the representation space, which is applied to local training on the user side.•We propose a label reconstruction strategy aimed at maximizing the utilization of noisy datasets by leveraging a semi-supervised learning scheme.

External IDs:dblp:journals/fgcs/LinJZFL25