Frequency-Aware Masked Autoencoders for Human Activity Recognition using Accelerometers
Open accessThe DOI matches its Crossref record; 21 of the 21 references it cites were found in a registry or an index.
Abstract
Wearable accelerometers are widely used for continuous monitoring of physical activity. Supervised machine learning and deep learning algorithms have long been used to extract meaningful activity information from raw accelerometry data, but progress has been hampered by the limited amount of labeled data that is publicly available. Exploiting large unlabeled datasets using self-supervised pretraining is a relatively new and underexplored approach in the field of human activity recognition (HAR).We used a time-series transformer masked autoencoder (MAE) approach to self-supervised pretraining and propose two novel spectrogram-based loss functions: the log-scale mean-magnitude (LMM) and log-scale magnitude variance (LMV) losses. We compared these losses with the mean squared error (MSE) loss for MAE training. We leveraged the large unlabeled UK Biobank accelerometry dataset (n = 109k) for pretraining and evaluated downstream HAR performance using a linear classifier in a smaller labelled dataset.We found that pretraining with the LMM loss improved performance compared to an MAE pretrained with the MSE loss, with 12.7% increase in subject-wise F1 score when using linear probing. Compared with a state-of-the-art ResNet-based HAR model, our LMM-pretrained transformer models performed better (+9.8% F1) with linear probing and comparably when fine-tuned using an LSTM classifier. The addition of the LMV to the LMM loss decreased performance compared to the LMM loss alone. These findings establish the LMM loss as a robust and effective method for pretraining MAE models on accelerometer data for HAR and show the potential of pretraining sequence-based models for free-living HAR.Clinical Relevance- Improved algorithms for human activity recognition, enables more accurate monitoring of physical activity, which is crucial for assessing mobility, rehabilitation progress, and chronic disease management.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Citations
Cites
Masked Autoencoders Are Scalable Vision Learners
He, K., Chen, X., Xie, S., Li, Y., Dollar, P., & Girshick, R. (2022) · 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52688.2022.01553 · Cited by 8,489 · cited 3×
“However, since the arrival of foundation models for images and time-series relying on self-supervised pretraining [13], [14], multiple studies have employed different methods of self-supervised learning to wearable accelerometry data [2], [15], [16], [1], [4], [5].”
Emerging Properties in Self-Supervised Vision Transformers
Caron, M., Touvron, H., Misra, I., Jegou, H., Mairal, J., Bojanowski, P., & Joulin, A. (2021) · 2021 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00951 · Cited by 5,638 · cited 1×
“This is to our knowledge the first demonstration of the viability of pretraining transformer MAEs on large-scale high-resolution accelerometry data for HAR.”
Doherty, A., Jackson, D., Hammerla, N., Plötz, T., Olivier, P., Granat, M. H., White, T., van Hees, V. T., Trenell, M. I., Owen, C. G., et al. (2017) · PLOS ONE · doi:10.1371/journal.pone.0169649 · Cited by 1,250 · cited 1×
“The unlabeled high-resolution wrist-worn accelerometer dataset from the UK Biobank [22] was used for self-supervised pretraining in this work.”
van Hees, V. T., Fang, Z., Langford, J., Assah, F., Mohammad, A., da Silva, I. C. M., Trenell, M. I., White, T., Wareham, N. J., & Brage, S. (2014) · Journal of Applied Physiology · doi:10.1152/japplphysiol.00421.2014 · Cited by 676 · cited 1×
“The accelerometry data was then calibrated using the algorithm described in [24] without temperature correction.”
Multi-task Self-Supervised Learning for Human Activity Detection
Saeed, A., Ozcelebi, T., & Lukkien, J. (2019) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3328932 · Cited by 346 · cited 2×
“Wearable accelerometry is a cost-effective tool for out-of-clinic health monitoring, with applications in human activity recognition (HAR) [1], [2], [3], [4], gait estimation [5], [6], and sleep monitoring [7].”
Root Mean Square Layer Normalization
Zhang, B., & Sennrich, R. (2019) · arXiv · doi:10.48550/arxiv.1910.07467 · Cited by 262 · cited 1×
“During finetuning, the full 24-hour recordings were loaded into memory and divided into 50-minute windows with 45 minutes of overlap.”
Willetts, M., Hollowell, S., Aslett, L., Holmes, C., & Doherty, A. (2018) · Scientific Reports · doi:10.1038/s41598-018-26174-1 · Cited by 250 · cited 3×
“The labels are reduced to the six categories bicycling, walking, mixed, vehicle, sit-stand, and sleep following previous research [1], [11].”
Physical Activity Classification Using the GENEA Wrist-Worn Accelerometer
ZHANG, S., ROWLANDS, A. V., MURRAY, P., & HURST, T. L. (2012) · Medicine & Science in Sports & Exercise · doi:10.1249/mss.0b013e31823bf95c · Cited by 233 · cited 1×
“Until recently, supervised classical machine learning and deep learning has been state-of-the-art for HAR using wearable accelerometers [8], [9], [10], [11], [12].”
Root Mean Square Layer Normalization
Zhang, B., & Sennrich, R. (2019) · Curran Associates · doi:10.5167/uzh-177483 · Cited by 203
Hip and Wrist Accelerometer Algorithms for Free-Living Behavior Classification
ELLIS, K., KERR, J., GODBOLE, S., STAUDENMAYER, J., & LANCKRIET, G. (2016) · Medicine & Science in Sports & Exercise · doi:10.1249/mss.0000000000000840 · Cited by 186 · cited 2×
“Until recently, supervised classical machine learning and deep learning has been state-of-the-art for HAR using wearable accelerometers [8], [9], [10], [11], [12].”
Cited by
Wang, J.-C., Chang, C.-T., & Ke, K. (2026) · The International Journal of Advanced Manufacturing Technology · doi:10.1007/s00170-026-17655-z · Cited by 1
UniMoRe : Uni fied Mo tion Re presentation via Gait IMU Pretraining
Chakraborty, S., Kumar, A., & Nigam, A. (2026) · IEEE Sensors Journal · doi:10.1109/jsen.2026.3718949
“Unimodal methods explore frequency-domain representations [9], [12], [13] or domain-informed objectives [7], [14], [15], [16].”
Metadata
- Authors
- Lorenzen, N. R., Jennum, P. J., Mignot, E., & Brink-Kjaer, A.
- Affiliations
- Technical University of Denmark (DK)
- Copenhagen University Hospital (DK)
- Stanford University (US)
- Technical University of Denmark,Department of Health Technology,Lyngby,Denmark
- Venue
- 2025 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)
- Publisher
- IEEE
- Published
- 2025-07-14
- Volume
- 2025
- Pages
- 1-7
- ISSN
- 2375-7477 (ISSN-L)
- OpenAlex
- W4416961249
- Cited by
- 2
- References
- 18
- Citations by year
- 2026: 2
- FWCI
- 1.49
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (publishedVersion)
- PubMed (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Context-Aware Activity Recognition Systems
- Health Sciences › Medicine › Public Health, Environmental and Occupational Health › Physical Activity and Health
- Physical Sciences › Computer Science › Signal Processing › Time Series Analysis and Forecasting
- Keywords
- human activity recognition, masked autoencoder, self-supervised pre-training, wearable accelerometers, time series transformer, linear probing, physical activity monitoring
- MeSH terms
- Machine Learning, Wearable Electronic Devices, Autoencoder, Algorithms, Humans, Human Activities, Accelerometry
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Frequency-Aware Masked Autoencoders for Human Activity Recognition using Accelerometers. https://app.cochat.ai/papers/10.1109/embc58623.2025.11252605
- The paper
- Lorenzen, N. R., Jennum, P. J., Mignot, E., & Brink-Kjaer, A. (2025). Frequency-Aware Masked Autoencoders for Human Activity Recognition using Accelerometers. 2025 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). https://doi.org/10.1109/embc58623.2025.11252605
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.