FedHAR: Semi-Supervised Online Learning for Personalized Federated Human Activity Recognition

Yu, Hongzheng; Chen, Zekai; Zhang, Xiao; Chen, Xu; Zhuang, Fuzhen; Xiong, Hui; Cheng, Xiuzhen

doi:10.1109/tmc.2021.3136853

Cited by 39 publications

(47 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This assumption comes along with the well-known drawbacks of supervised learning on sensor data, e.g., the challenges associated with data labeling. We note that unsupervised [79] and semi-supervised FL [3,62,77] are active research areas in ML and future works can investigate them in the context of multi-device systems.…”

Section: Discussion and Limitationsmentioning

confidence: 99%

“…One of the key motivations behind FL is that local clients have an insufficient amount of training data to learn a good prediction model, and hence they collaborate with other clients to learn a shared prediction model. Unfortunately, some of the prior work on FL with HAR data disregards this assumption; for example, Yu et al [77] assume between 3000 to 5000 seconds of labeled data on each client for the RealWorld dataset. This raises two concerns: firstly, it is infeasible for users to label several hours of accelerometer and gyroscope data on their devices.…”

Section: Experiments Testbedmentioning

confidence: 99%

“…Further, we compare FLAME against Oort [37] which is an extension of FedAvg with heterogeneity-aware device selection policies. As our work operates under the paradigm of supervised FL, we do not compare it against any unsupervised or semi-supervised FL baselines [2,77].…”

Section: Baselinesmentioning

confidence: 99%

“…FedDL [71] and FjORD [25] dynamically adapt model layers and model size for heterogeneous devices to participate in FL. Some studies further employ self-supervised or semi-supervised learning methods to utilize unlabeled sensor data for personalized federated learning [3,62,77]. Meta-HAR [40] adopts Model-Agnostic Meta Learning to boost the representation ability of the shared embedding network and uses personalization for adaptation on top of it.…”

Section: Related Workmentioning

confidence: 99%

“…A Realistic Testbed for FL. Prior works on federated training of HAR models (e.g., [77]) have assumed that each FL client owns a large amount of labeled data, in the order of 3000-5000 seconds. Instead, we setup a realistic FL testbed with a large number of clients, each holding a small amount of labeled data.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

FLAME: Federated Learning Across Multi-device Environments

Cho¹,

Mathur²,

Kawsar³

2022

Preprint

View full text Add to dashboard Cite

Federated Learning (FL) enables distributed training of machine learning models while keeping personal data on user devices private. While we witness increasing applications of FL in the area of mobile sensing, such as human-activity recognition, FL has not been studied in the context of a multi-device environment (MDE), wherein each user owns multiple data-producing devices. With the proliferation of mobile and wearable devices, MDEs are increasingly becoming popular in ubicomp settings, therefore necessitating the study of FL in them. FL in MDEs is characterized by high non-IID-ness across clients, complicated by the presence of both user and device heterogeneities. Further, ensuring efficient utilization of system resources on FL clients in a MDE remains an important challenge. In this paper, we propose FLAME, a user-centered FL training approach to counter statistical and system heterogeneity in MDEs, and bring consistency in inference performance across devices. FLAME features (i) user-centered FL training utilizing the time alignment across devices from the same user; (ii) accuracy-and efficiency-aware device selection; and (iii) model personalization to devices. We also present an FL evaluation testbed with realistic energy drain and network bandwidth profiles, and a novel class-based data partitioning scheme to extend existing HAR datasets to a federated setup. Our experiment results on three multi-device HAR datasets show that FLAME outperforms various baselines by 4.8-33.8% higher 𝐹 1 score, 1.02-2.86× greater energy efficiency, and up to 2.02× speedup in convergence to target accuracy through fair distribution of the FL workload.CCS Concepts: • Human-centered computing → Ubiquitous and mobile computing systems and tools; • Computing methodologies → Cooperation and coordination.

show abstract

Section: Discussion and Limitationsmentioning

confidence: 99%

Section: Experiments Testbedmentioning

confidence: 99%

Section: Baselinesmentioning

confidence: 99%