Paperdoi:10.1145/3544794.3558467

TinyHAR: A Lightweight Deep Learning Model Designed for Human Activity Recognition

Venue listed2022Added Nov 12, 2026

  • YZ
  • HZ
  • YH
  • TR
  • MH
  • MK

Authors: Yexu Zhou, Haibin Zhao, Yiran Huang, Till Riedel, Michael Hefenbrock, Michael Beigl (KIT)

ISWC '22 (ACM International Symposium on Wearable Computers)

TL;DR

A small HAR architecture that combines 1D convolutions, a transformer-style cross-channel encoder, and a lightweight attention / recurrent tail, aimed at wearable budgets rather than server TAL models. The official repo is the ISWC 2022 HAR code release. Later work measures its size and on-device cost more carefully than the original paper's headline tables.

  • Form factor: multi

Abstract

Deep learning models have shown excellent performance in human activity recognition tasks. However, these models typically require large amounts of computational resources, which makes them inefficient to deploy on edge devices. Furthermore, the superior performance of deep learning models relies heavily on the availability of large datasets to avoid over-fitting. However, the expensive efforts for labeling limits the amount of datasets. We address both challenges by designing a more lightweight model, called TinyHAR. TinyHAR is designed specifically for human activity recognition employing different saliency of multi modalities, multimodal collaboration, and temporal information extraction. Initial experimental results show that TinyHAR is several times smaller and often meets or even surpasses the performance of DeepConvLSTM, a state-of-the-art human activity recognition model.

Verbatim from Semantic Scholar (opens in a new tab) · Abstract via Semantic Scholar; © the authors/publisher

Results on ChatPPG leaderboards

All leaderboards

No results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.

Tasks & signals

Tasks

Activity recognition

Signals

Keywords

HarTinymlTinyharEdgeIswc 2022

Datasets used

How it works

How Method works: Setup → Method → TargetSetup: Signals (accel, gyro); Form factor (multi). Method: Method. Target: Activity recognition.SetupSignals: accel, gyroxyzSignalsaccel, gyroForm factor: multiForm factormultiMethodMethodTargetActivity recognitionActivity recognition

Cite this paper

Cite this paper with the fields recorded here. Check it against the publisher before submitting.

Zhou, Y., Zhao, H., Huang, Y., Riedel, T., Hefenbrock, M., & Beigl, M. (2022). TinyHAR: A Lightweight Deep Learning Model Designed for Human Activity Recognition. ISWC '22. https://doi.org/10.1145/3544794.3558467
Zhou, Yexu, Haibin Zhao, Yiran Huang, Till Riedel, Michael Hefenbrock, and Michael Beigl. 2022. "TinyHAR: A Lightweight Deep Learning Model Designed for Human Activity Recognition." ISWC '22. https://doi.org/10.1145/3544794.3558467.
Zhou, Y., Zhao, H., Huang, Y., Riedel, T., Hefenbrock, M. and Beigl, M. (2022) 'TinyHAR: A Lightweight Deep Learning Model Designed for Human Activity Recognition', ISWC '22. doi: 10.1145/3544794.3558467.
Zhou, Yexu, et al. "TinyHAR: A Lightweight Deep Learning Model Designed for Human Activity Recognition." ISWC '22, 2022. https://doi.org/10.1145/3544794.3558467.
Y. Zhou, H. Zhao, Y. Huang, T. Riedel, M. Hefenbrock, and M. Beigl, "TinyHAR: A Lightweight Deep Learning Model Designed for Human Activity Recognition," ISWC '22, 2022, doi: 10.1145/3544794.3558467.
@inproceedings{zhou2022tinyhar,
  title         = {{TinyHAR: A Lightweight Deep Learning Model Designed for Human Activity Recognition}},
  author        = {Zhou, Yexu and Zhao, Haibin and Huang, Yiran and Riedel, Till and Hefenbrock, Michael and Beigl, Michael},
  year          = {2022},
  booktitle     = {ISWC '22},
  doi           = {10.1145/3544794.3558467}
}

Added to the ChatPPG research hub