Human Activity Recognition Using Federated Learning
Abstract
State-of-the-art deep learning models for human activity recognition use large amount of sensor data to achieve high accuracy. However, training of such models in a data center using data collected from smart devices leads to high communication costs and possible privacy infringement. In order to mitigate aforementioned issues, federated learning can be employed to train a generic classifier by combining multiple local models trained on data originating from multiple clients. In this work we evaluate federated learning to train a human activity recognition classifier and compare its performance to centralized learning by building two models, namely a deep neural network and a softmax regression trained on both synthetic and real-world datasets. We study communication costs as well as the influence of erroneous clients with corrupted data in federated learning setting. We have found that federated learning for the task of human activity recognition is capable of producing models with slightly worse, but acceptable, accuracy compared to centralized models. In our experiments federated learning achieved an accuracy of up to 89% compared to 93% in centralized training for the deep neural network. The global model trained with federated learning on skewed datasets achieves accuracy comparable to centralized learning. Furthermore, we identified an important issue of clients with corrupted data and proposed a federated learning algorithm that identifies and rejects erroneous clients. Lastly, we have identified a trade-off between communication cost and the complexity of a model. We show that more complex models such as deep neural network require more communication in federated learning settings for human activity recognition compared to less complex models, such as multinomial logistic regression.
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?
Federated learning is a trending approach to resolve privacy issues in learning problems.
“Federated learning is a trending approach to resolve privacy issues in learning problems [275,276,277,278].”
From Deep Learning in Human Activity Recognition with Wearable Sensors: A Review on Advances (2022) · cites this paper as [277]
Arivazhagan et al. propose a neural network architecture where global data is used to train only the base layers, while the personalization layers are trained on local data.
“Arivazhagan et al. [134] propose a neural network architecture, where global data is used to train only the base layers, while the personalization layers are trained on local data.”
From Challenges, Applications and Design Aspects of Federated Learning: A Survey (2021) · cites this paper as [134]
The FEDPER approach outperforms a pure FedAvg approach using the Flickr-AES dataset.
“The authors empirically show that the FEDPER approach outperforms a pure FedAvg approach using the Flickr-AES dataset [134] considering the personalization layers are capable of representing the personal predilection of an FL user.”
From Challenges, Applications and Design Aspects of Federated Learning: A Survey (2021) · cites this paper as [134]
FL requires less communication and computational resources while using less complex models, but at the cost of lower accuracy.
“Then, they study the trade-off between the communication cost and the model accuracy, which indicates that FL requires less communication and computational resources while using less complex models but at the cost of lower accuracy.”
From Handling privacy-sensitive medical data with federated learning: Challenges and future directions (2023) · cites this paper as [37]
FL requires less communication overhead, and hence it can replace traditional learning approaches in IoMT applications.
“It can be learned from [37] that FL requires less communication overhead, and hence it can replace traditional learning approaches in IoMT applications.”
From Handling privacy-sensitive medical data with federated learning: Challenges and future directions (2023) · cites this paper as [37]
When the training data is IID, the difference between FL and non-FL is within 3%.
“Also, the authors showed that when the training data is IID, the difference between FL and non-FL is within 3%.”
From Handling privacy-sensitive medical data with federated learning: Challenges and future directions (2023) · cites this paper as [37]
Using less complex models such as softmax regression is a viable solution for most real-world applications because user behavior phenomena are often simple enough to be captured by relatively simple models.
“Results confirm that using less complex models, such as softmax regression, is a viable solution for most real-world applications since user behavior phenomena are often simple enough to be captured by relatively simple models.”
From Handling privacy-sensitive medical data with federated learning: Challenges and future directions (2023) · cites this paper as [37]
The citation lists haven’t been fetched yet
References 19 · Cited by 167
- Authors
- Sozinov, K., Vlassov, V., & Girdzijauskas, S.
- Affiliations
- KTH Royal Institute of Technology (SE)
- Venue
- 2018 IEEE Intl Conf on Parallel & Distributed Processing with Applications, Ubiquitous Computing & Communications, Big Data & Cloud Computing, Social Computing & Networking, Sustainable Computing & Communications (ISPA/IUCC/BDCloud/SocialCom/SustainCom)
- Publisher
- IEEE
- Published
- 2018-12
- Pages
- 1103-1111
- OpenAlex
- W2922569945
- Cited by
- 167
- References
- 19
- Citations by year
- 2026: 8 · 2025: 26 · 2024: 36 · 2023: 28 · 2022: 27 · 2021: 29 · 2020: 13
- FWCI
- 4.29
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- IEEE International Symposium on Parallel and Distributed Processing with Applications (ISPA) (publishedVersion)
- 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
- human activity recognition, federated learning, communication overhead, privacy preservation, deep neural network, softmax regression, multinomial logistic regression, model complexity
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
