FedHAR: Semi-Supervised Online Learning for Personalized Federated Human Activity Recognition
Abstract
The advancement of smartphone sensors and wearable devices has enabled a new paradigm for smart human activity recognition (HAR), which has a broad range of applications in healthcare and smart cities. However, there are four challenges,privacy preservation,label scarcity,real-timing, andheterogeneity patterns, to be addressed before HAR can be more applicable in real-world scenarios. To this end, in this paper, we propose a personalized federated HAR framework, namedFedHAR, to overcome all the above obstacles. Specially, as federated learning,FedHARperforms distributed learning, which allows training data to be kept local to protect users’ privacy. Also, for each client without activity labels, inFedHAR, we design an algorithm to compute unsupervised gradients under theconsistency trainingproposition and an unsupervised gradient aggregation strategy is developed for overcoming the concept drift and convergence instability issues in online federated learning process. Finally, extensive experiments are conducted using two diverse real-world HAR datasets to show the advantages ofFedHARover state-of-the-art methods. In addition, when fine-tuning each unlabeled client, personalizedFedHARcan achieve additional 10% improvement across all metrics on average.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Personalized HAR frameworks typically adopt a two-stage approach: pre-training on a general dataset followed by fine-tuning on data from the target user.
“Prior works have explored personalized HAR frameworks [11,12,13,14,15,16], typically adopting a two-stage approach: pre-training on a general dataset followed by fine-tuning on data from the target user.”
From Towards User-Generalizable Wearable-Sensor-Based Human Activity Recognition: A Multi-Task Contrastive Learning Approach (2025) · cites this paper as [16]
Personalized HAR methods suffer from increased deployment costs because collecting data from each new user is required.
“While effective, these methods suffer from two major limitations: (1) collecting data from each new user increases deployment costs, and (2) the two-stage training pipeline adds complexity and computational overhead.”
From Towards User-Generalizable Wearable-Sensor-Based Human Activity Recognition: A Multi-Task Contrastive Learning Approach (2025) · cites this paper as [16]
Personalized HAR methods suffer from added complexity and computational overhead due to the two-stage training pipeline.
“While effective, these methods suffer from two major limitations: (1) collecting data from each new user increases deployment costs, and (2) the two-stage training pipeline adds complexity and computational overhead.”
From Towards User-Generalizable Wearable-Sensor-Based Human Activity Recognition: A Multi-Task Contrastive Learning Approach (2025) · cites this paper as [16]
ProtoHAR and FedHAR still require new user data to be collected and labeled, followed by local re-training, which is a two-stage process.
“However, they still require new user data to be collected and labeled, followed by local re-training—again, a two-stage process.”
From Towards User-Generalizable Wearable-Sensor-Based Human Activity Recognition: A Multi-Task Contrastive Learning Approach (2025) · cites this paper as [16]
ProtoHAR and FedHAR employed federated learning to build user-specific models.
“We also compared our method with personalization-based approaches such as ProtoHAR [44] and FedHAR [16], which employed federated learning to build user-specific models.”
From Towards User-Generalizable Wearable-Sensor-Based Human Activity Recognition: A Multi-Task Contrastive Learning Approach (2025) · cites this paper as [16]
The federated learning design of ProtoHAR and FedHAR effectively addresses privacy concerns by keeping user data local.
“This design effectively addresses privacy concerns by keeping user data local.”
From Towards User-Generalizable Wearable-Sensor-Based Human Activity Recognition: A Multi-Task Contrastive Learning Approach (2025) · cites this paper as [16]
ProtoHAR and FedHAR first trained a general model and then fine-tuned personalized models on client devices.
“Both methods first trained a general model and then fine-tuned personalized models on client devices.”
From Towards User-Generalizable Wearable-Sensor-Based Human Activity Recognition: A Multi-Task Contrastive Learning Approach (2025) · cites this paper as [16]
The citation lists haven’t been fetched yet
References 45 · Cited by 113
- Authors
- Yu, H., Chen, Z., Zhang, X., Chen, X., Zhuang, F., Xiong, H., & Cheng, X.
- Affiliations
- Shandong University (CN)
- George Washington University (US)
- Sun Yat-sen University (CN)
- Beihang University (CN)
- Hong Kong University of Science and Technology (HK)
- Venue
- IEEE Transactions on Mobile Computing
- Published
- 2023-06-01
- Volume
- 22
- Issue
- 6
- Pages
- 3318-3332
- ISSN
- 1536-1233 (ISSN-L)
- 1558-0660
- 2161-9875
- OpenAlex
- W4205335850
- Cited by
- 113
- References
- 45
- Citations by year
- 2026: 16 · 2025: 34 · 2024: 27 · 2023: 30 · 2022: 6
- FWCI
- 5.57
- Type
- journal article
- Language
- en
- Open access
- closed
- Licence
- Locations
- IEEE Transactions on Mobile Computing (publishedVersion)
- Rare & Special e-Zone (The Hong Kong University of Science and Technology) (submittedVersion)
- Funders
- National Key Research and Development Program of China (2021ZD0113602)
- National Natural Science Foundation of China (62176014, U20A20159, U1711265, 61972432)
- Shandong Provincial Natural Science Foundation of China (ZR2021QF044)
- Program for Guangdong Introducing Innovative and Entrepreneurial Teams (2017ZT07X355)
- Guangdong Provincial Pearl River Talents Program (2017GC010465)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Context-Aware Activity Recognition Systems
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Keywords
- federated learning, human activity recognition, personalized federated learning, semi-supervised learning, online learning, consistency regularization, concept drift, privacy preservation, wearable sensors, smartphone sensors, limited labeled data
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
