Contrastive Predictive Coding for Human Activity Recognition
Open accessThe DOI matches its Crossref record; 117 of the 117 references it cites were found in a registry or an index; 20 of the 41 sources its claims cite were checked for support.
Abstract
Feature extraction is crucial for human activity recognition (HAR) using body-worn movement sensors. Recently, learned representations have been used successfully, offering promising alternatives to manually engineered features. Our work focuses on effective use of small amounts of labeled data and the opportunistic exploitation of unlabeled data that are straightforward to collect in mobile and ubiquitous computing scenarios. We hypothesize and demonstrate that explicitly considering the temporality of sensor data at representation level plays an important role for effective HAR in challenging scenarios. We introduce the Contrastive Predictive Coding (CPC) framework to human activity recognition, which captures the temporal structure of sensor data streams. Through a range of experimental evaluations on real-life recognition tasks, we demonstrate its effectiveness for improved HAR. CPC-based pre-training is self-supervised, and the resulting learned representations can be integrated into standard activity chains. It leads to significantly improved recognition performance when only small amounts of labeled training data are available, thereby demonstrating the practical value of our approach. Through a series of experiments, we also develop guidelines to help practitioners adapt and modify the framework towards other mobile and ubiquitous computing scenarios.
Overview
CoChat Summary
TL;DR
The paper introduces Contrastive Predictive Coding (CPC) as self-supervised pre-training for wearable human activity recognition, learning representations by predicting multiple future timesteps in latent space. Across four benchmark datasets, CPC-based frozen features outperform prior unsupervised methods on three datasets and improve recognition when only small amounts of labeled data are available.
Key findings
- CPC unsupervised mean F1-scores were Mobiact 80.97, Motionsense 89.05, UCI-HAR 81.65, and USC-HAD 52.01; it outperformed all state-of-the-art unsupervised baselines on Mobiact, Motionsense, and USC-HAD, and was comparable to masked reconstruction on UCI-HAR (81.89).
- Relative to supervised DeepConvLSTM, CPC improved performance by approximately 4% on Motionsense and over 7% on USC-HAD; on Mobiact and UCI-HAR, CPC scored below DeepConvLSTM in Table 2 (80.97 vs 82.40; 81.65 vs 82.83).
- With limited labeled data, CPC improved over DeepConvLSTM on all datasets: Mobiact was consistently higher by over 5% and over 15% above random-initialized features when more than ten samples per class were available; Motionsense reached a maximum improvement of approximately 10% at 50 samples; UCI-HAR peaked at over 10% at ten samples; USC-HAD improved even with one sample per class.
- For pre-training choices, k=12 future steps was best for Mobiact, Motionsense, and UCI-HAR (about 400 ms), while k=8 was best for USC-HAD (about 270 ms); kernel size 9 was around 13% lower than kernel size 3, recurrent encoders were around 42% lower, and using only the first two encoder layers gave Mobiact 85.22% and UCI-HAR 82.58%.
Methods
- Self-supervised CPC pre-training on unlabeled sliding windows: an encoder \(g_{enc}\) maps sensor windows to latent representations, a GRU autoregressive network \(g_{ar}\) produces context representations, and linear predictors \(W_k\) predict k future timesteps using the InfoNCE/NCE loss. Frozen learned weights are used as a feature extractor with a three-layer MLP classifier (256, 128, softmax; batch normalization, ReLU, dropout p=0.2).
- Four benchmark datasets were used: Mobiact (61 users, 11 activities), Motionsense (24 users, 6 activities), UCI-HAR (30 users, 6 activities), and USC-HAD (14 users, 12 activities). Raw accelerometer and gyroscope data were downsampled to 30 Hz, normalized, and segmented into 1 s windows with 50% overlap; splits were participant-based, with 20% test and 20% of remaining data for validation, except USC-HAD used participants 1–10 for training, 11–12 for validation, and 13–14 for testing.
- Pre-training ran for 150 epochs with Adam, learning rates tuned over {1e-3, 5e-4}, and k over {2, 4, 8, 12, 16}; the classifier was trained for 150 epochs with cross-entropy, learning rates over {5e-4, 1e-4} decayed by 0.8 every 25 epochs. The primary metric was test set mean F1-score.
Limitations
- The paper does not state a dedicated limitations section. Evident from the text: evaluation is limited to four benchmark datasets and primarily locomotion-style activities, and the authors note open challenges for sensor placements different from the source dataset and domain shift when using virtual movement data.
- Evident from the text: CPC pre-training uses the entire unlabeled dataset for each benchmark, and the paper does not report cross-dataset transfer or deployment on new sensor placements; it frames these as future research agenda.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
14 Valid · 7 Not valid · 7 Not checked · 84 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“We introduce the Contrastive Predictive Coding (CPC) framework to human activity recognition, which captures the long-term temporal structure of sensor data streams.”
What the sources say
- Partly supportedcites [58] [58] Aaron van den Oord, Yazhe Li, and Oriol Vinyals. 2018. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748 (2018). sourceExperimental — accuracy not yet measured
Section 2.4 (p. 4) says CPC extracts features that retain information across long time horizons, supporting the temporal-structure part of the claim. The paper tests audio, images, language, and reinforcement learning (Conclusion, p. 9), but does not establish that CPC captures long-term structure in the sensor data streams used for human activity recognition. (LLM, conf=0.84)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“Therefore, it is applicable to a wider range of scenarios and environments.”
What the sources say
- Could not verifycites [69] [69] Aaqib Saeed, Tanir Ozcelebi, and Johan Lukkien. 2019. Multi-task Self-Supervised Learning for Human Activity Detection. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 3, 2 (2019), 1-30. - sourceExperimental — accuracy not yet measured
Section 4.1 (p. 8) says the source evaluated its method across datasets covering different devices and environments. The paper also says its approach could apply to a wide variety of problems (p. 1). Neither passage establishes whether the citing paper's work applies to a wider range of scenarios and environments than this method. (LLM, conf=0.76)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
Every source the paper cites for this claim was checked and supports it.
In the paper
“This results in representations that encode the high-level information between temporally separated parts of the time-series signal [58], thereby leading to improved downstream recognition performance.”
What the sources say
- Supportedcites [58] [58] Aaron van den Oord, Yazhe Li, and Oriol Vinyals. 2018. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748 (2018). source
Section 2.1 (p. 2) says CPC learns representations of information shared between different parts of a signal while discarding local detail; predicting further into the future requires the model to infer more global structure. Section 2.4 (p. 4) describes features spanning long time horizons, supporting the paraphrase about high-level information in temporally separated parts of a time series. (LLM, conf=0.93)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“While [32] utilizes both accelerometer and gyroscope data, it delivers mixed results relative to DeepConvLSTM [59] when fine-tuning the pre-trained weights using limited labeled data.”
What the sources say
- Not matchedcites [32] [32] Harish Haresamudram, Apoorva Beedu, Varun Agrawal, Patrick L Grady, Irfan Essa, Judy Hoffman, and Thomas Plötz. 2020. Masked reconstruction based self-supervision for human activity recognition. In Proceedings of the 2020 International Symposium on Wearable Computers . 45-49. -
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Supportedcites [59] [59] Francisco Javier Ordóñez and Daniel Roggen. 2016. Deep convolutional and lstm recurrent neural networks for multimodal wearable activity recognition. Sensors 16, 1 (2016), 115. - source
The citation to [59] identifies DeepConvLSTM as the comparison model; it does not attribute the Transformer fine-tuning results to this source. Section 3 (Architecture) introduces DeepConvLSTM, confirming that identification. (LLM, conf=0.92)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Not matched
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“While they result in effective representations, most of these approaches do not specifically target a characteristic inherent to body-worn sensor data - temporality.”
What the sources say
- Partly supportedcites [31] [31] Harish Haresamudram, David V Anderson, and Thomas Plötz. 2019. On the role of features in human activity recognition. In Proceedings of the 23rd International Symposium on Wearable Computers . 78-88. - sourceExperimental — accuracy not yet measured
Section 2 (pp. 1–2) describes body-worn sensor data as time series and says movement-data frames are usually treated as statistically independent. It also discusses pre-trained networks, but does not establish that most unsupervised pre-training approaches fail to target temporality; its account of their results is mixed rather than uniformly effective. (LLM, conf=0.78)
- Not checked — no full text of the sourcecites [74] [74] Alireza Abedin Varamin, Ehsan Abbasnejad, Qinfeng Shi, Damith C Ranasinghe, and Hamid Rezatofighi. 2018. Deep auto-set: A deep auto-encoder-set network for activity recognition using wearables. In Proceedings of the 15th EAI International Conference on Mobile and Ubiquitous Systems: Computing, Networking and Services . 246-253. - source
- Not matchedcites [32] [32] Harish Haresamudram, Apoorva Beedu, Varun Agrawal, Patrick L Grady, Irfan Essa, Judy Hoffman, and Thomas Plötz. 2020. Masked reconstruction based self-supervision for human activity recognition. In Proceedings of the 2020 International Symposium on Wearable Computers . 45-49. -
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Partly supportedcites [69] [69] Aaqib Saeed, Tanir Ozcelebi, and Johan Lukkien. 2019. Multi-task Self-Supervised Learning for Human Activity Detection. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 3, 2 (2019), 1-30. - sourceExperimental — accuracy not yet measured
The Related Work section (pp. 22–23) treats sensor activity data as temporal and distinguishes the authors’ transformation-based method from earlier unsupervised pre-training methods. It does not say that most such approaches fail to target temporality; that comparative claim is unsupported by our copy. (LLM, conf=0.78)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
77 more claims · 2 with their cited sources checked
Create a free account to see all 112 claims and to verify the 4 references still open.
- Probably not a substantial claim. 21
- It may state two separate claims as one. 15
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?
The cited source doesn’t support the claim as written.
In the citing paper
“Leave-one-group-out cross-validation approach is used for performance evaluation follows the previous works [26,34] .”
From Spatial-Temporal Masked Autoencoder for Multi-Device Wearable Human Activity Recognition (2023) · cites this paper as [26]
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Both methods were designed to leverage unlabeled data by pre-training on unannotated signals and subsequently fine-tuning on labeled data.”
From Towards User-Generalizable Wearable-Sensor-Based Human Activity Recognition: A Multi-Task Contrastive Learning Approach (2025) · cites this paper as [30]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“However, the contrastive loss function of the above work is InfoNCE [39] or NT-Xent [40]; both are essentially the instance discrimination task that will regard the same-class samples as negative examples.”
From Negative Selection by Clustering for Contrastive Learning in Human Activity Recognition (2023) · cites this paper as [39]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“In contrast, previous studies applied self-supervised training and fine-tuning only on the same data sources24,27,40, making it necessary to pre-train on every new dataset in practical applications.”
From Self-supervised learning for human activity recognition using 700,000 person-days of wearable data (2024) · cites this paper as [40]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Recently a number of contributions have been made that introduced self-supervised learning into the field of HAR, including, Multi-task self-supervision, Masked Reconstruction, CPC, and SimCLR, to name but a few.”
From Assessing the State of Self-Supervised Human Activity Recognition Using Wearables (2022) · cites this paper as [27]
Experimental — accuracy not yet measured
12 more claims about this paper
Create a free account to see every claim other papers make about this one.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 6.1 Analyzing Contrastive Predictive Coding · p. 16
“As mentioned in [58], either 𝑐 𝑡 (the output of 𝑔 𝑒𝑛𝑐 ) or 𝑧 𝑡 (the output of 𝑔 𝑎𝑟 ) can be utilized as the representation.”
§ 3.1 Contrastive Predictive Coding · p. 6
“a non-linear encoder 𝑔 𝑒𝑛𝑐 is utilized to map a window of sensor data 𝑥 𝑡 to a sequence of latent representations given by 𝑧 𝑡 = 𝑔 𝑒𝑛𝑐 ( 𝑥 𝑡 ) . Subsequently, an autoregressive model 𝑔 𝑎𝑟 is used to summarize all 𝑧 ≤ 𝑡 into a latent space, providing a context latent representation 𝑐 𝑡 = 𝑔 𝑎𝑟 ( 𝑧 ≤ 𝑡 ) .”
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 1×
“The availability of large-scale annotated datasets has resulted in astonishing improvements in performance due to the application of deep learning to computer vision [34, 42], speech recognition [3, 26] and natural language tasks [18, 52].”
BackgroundVaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155 · cited 1×
“The availability of large-scale annotated datasets has resulted in astonishing improvements in performance due to the application of deep learning to computer vision [34, 42], speech recognition [3, 26] and natural language tasks [18, 52].”
BackgroundHochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318 · cited 1×
“[59] where 2D convolutional features are fed to a long short-term memory [36] network to perform activity recognition.”
BackgroundAdam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946
LeCun, Y., Bengio, Y., & Hinton, G. (2015) · Nature · doi:10.1038/nature14539 · Cited by 84,126
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173
Glove: Global Vectors for Word Representation
Pennington, J., Socher, R., & Manning, C. (2014) · Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.3115/v1/d14-1162 · Cited by 34,071 · cited 1×
“In the case of natural language processing, context-based self-supervised learning has been applied to obtain word and sentence level embeddings for approaches such as Word2Vec [52], GloVe [62] and universal sentence encoder [9].”
Method(2014) · Cited by 33,419
PyTorch: An Imperative Style, High-Performance Deep Learning Library
Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al. (2019) · arXiv · doi:10.48550/arxiv.1912.01703 · Cited by 31,473 · cited 1×
“We implemented our framework using PyTorch [60].”
Research works printed but not linked: 4
- 10.5555/1460232.1460247 · doi:10.5555/1460232.1460247
- William Lotter , Gabriel Kreiman , and David Cox . 2016. Deep predictive coding networks for video prediction and unsupervised learning. arXiv preprint arXiv:1605.08104 ( 2016 ). William Lotter, Gabriel Kreiman, and David Cox. 2016. Deep predictive coding networks for video prediction and unsupervised learning. arXiv preprint arXiv:1605.08104 (2016). · arXiv
- Aaqib Saeed , Victor Ungureanu , and Beat Gfeller . 2020. Sense and Learn: Self-Supervision for Omnipresent Sensors. arXiv preprint arXiv:2009.13233 ( 2020 ). Aaqib Saeed, Victor Ungureanu, and Beat Gfeller. 2020. Sense and Learn: Self-Supervision for Omnipresent Sensors. arXiv preprint arXiv:2009.13233 (2020). · arXiv
- Neo Wu , Bradley Green , Xue Ben , and Shawn O'Banion . 2020. Deep transformer models for time series forecasting: The influenza prevalence case. arXiv preprint arXiv:2001.08317 ( 2020 ). Neo Wu, Bradley Green, Xue Ben, and Shawn O'Banion. 2020. Deep transformer models for time series forecasting: The influenza prevalence case. arXiv preprint arXiv:2001.08317 (2020). · arXiv
Cited by
Self-supervised learning for human activity recognition using 700,000 person-days of wearable data
Yuan, H., Chan, S., Creagh, A. P., Tong, C., Acquah, A., Clifton, D. A., & Doherty, A. (2024) · npj Digital Medicine · doi:10.1038/s41746-024-01062-3 · Cited by 166
“, 2020a), contrastive learning (Haresamudram et al., 2020b; Tang et al., 2020), and multi-task SSL (Saeed et al.”
Human Activity Recognition: Review, Taxonomy and Open Challenges
Arshad, M. H., Bilal, M., & Gani, A. (2022) · Sensors · doi:10.3390/s22176463 · Cited by 206
“[27] Proposed a CPC framework based on CNN and LSTM for monitoring construction equipment activity.”
BackgroundMethodHuman activity recognition using wearable sensors by heterogeneous convolutional neural networks
Han, C., Zhang, L., Tang, Y., Huang, W., Min, F., & He, J. (2022) · Expert Systems with Applications · doi:10.1016/j.eswa.2022.116764 · Cited by 130
ColloSSL: Collaborative Self-Supervised Learning for Human Activity Recognition
Jain, Y., Tang, C. I., Min, C., Kawsar, F., & Mathur, A. (2022) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3517246 · Cited by 113
“[22] also proposed Contrastive Predictive Coding which leverages the long-term temporal structure of sensor data streams for self-supervised learning.”
BackgroundAssessing the State of Self-Supervised Human Activity Recognition Using Wearables
Haresamudram, H., Essa, I., & Plötz, T. (2022) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3550299 · Cited by 107
“The solved tasks need to be replaced with more challenging and relevant ones, addressing discovered flaws in the established experimental setups [17].”
BackgroundMethodInfluentialSelf-citationCosmo: contrastive fusion learning with small data for multimodal human activity recognition
Ouyang, X., Shuai, X., Zhou, J., Shi, I. W., Xie, Z., Xing, G., & Huang, J. (2022) · Proceedings of the 28th Annual International Conference on Mobile Computing And Networking · doi:10.1145/3495243.3560519 · Cited by 94
“There are also contrastive predictive coding methods designed for single-modality HAR applications that capture the temporal structure of sensor data [19, 48, 54].”
BackgroundMethodInfluentialWearable Sensor-Based Human Activity Recognition with Hybrid Deep Learning Model
Luwe, Y. J., Lee, C. P., & Lim, K. M. (2022) · Informatics · doi:10.3390/informatics9030056 · Cited by 92
“It can be summarized that the accuracy of deep learning techniques is sorted from low to high as 79.86%, 82.50%, 83.30%, 85.59%, 85.75%, 89.00% and 89.08% of BERT [28], CAE [29], multi-task self-supervision CNN [30], DeepConvLSTM [31], CNN [32], DCNN [33] and CPC [34], in ascending order.”
BackgroundMethodInfluentialCOCOA: Cross Modality Contrastive Learning for Sensor Data
Deldari, S., Xue, H., Saeed, A., Smith, D. V., & Salim, F. D. (2022) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3550316 · Cited by 92
“human activity recognition, sleep stage detection and emotion recognition) while existing rival methods only consider single type of sensors (mostly accelerometer data [36, 42]).”
BackgroundMethodInfluentialContrastive Self-supervised Learning for Sensor-based Human Activity Recognition
Khaertdinov, B., Ghaleb, E., & Asteriadis, S. (2021) · 2021 IEEE International Joint Conference on Biometrics (IJCB) · doi:10.1109/ijcb52358.2021.9484410 · Cited by 95
“We compare the performance of our SSL model in terms of mean F1-score against models reported in [8].”
BackgroundMethodResultInfluentialYin, Y., Xie, L., Jiang, Z., Xiao, F., Cao, J., & Lu, S. (2024) · IEEE Communications Surveys & Tutorials · doi:10.1109/comst.2024.3357591 · Cited by 108
“When combining multi-modal sensor data, different recognition approaches were also used [135], e.g., using image processing to track fingers in 2D space, while using phase changes of ultrasound to track the depth of fingers in 3D space, to provide a depth-aware tapping scheme.”
Sensor Data Augmentation by Resampling in Contrastive Learning for Human Activity Recognition
Wang, J., Zhu, T., Gan, J., Chen, L. L., Ning, H., & Wan, Y. (2022) · IEEE Sensors Journal · doi:10.1109/jsen.2022.3214198 · Cited by 76
Hong, Z., Li, Z., Zhong, S., Lyu, W., Wang, H., Ding, Y., He, T., & Zhang, D. (2024) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3659597 · Cited by 67
“Recently, with the increase of large-scale unlabeled data, weak-supervised, self-supervised, and unsupervised methods are receiving more attention [22, 23, 29, 58, 62, 68, 83, 86].”
BackgroundHarmony: Heterogeneous Multi-Modal Federated Learning through Disentangled Model Training
Ouyang, X., Xie, Z., Fu, H., Cheng, S., Pan, L., Ling, N., Xing, G., Zhou, J., & Huang, J. (2023) · International Conference on Mobile Systems, Applications, and Services · doi:10.1145/3581791.3596844 · Cited by 109
“the model performance can be augmented by efficiently learning from the unlabeled multi-modal [42] or unimodal [23, 61] data.”
BackgroundTwo-stream transformer network for sensor-based human activity recognition
Xiao, S., Wang, S., Huang, Z., Wang, Y., & Jiang, H. (2022) · Neurocomputing · doi:10.1016/j.neucom.2022.09.099 · Cited by 75
IMUGPT 2.0: Language-Based Cross Modality Transfer for Sensor-Based Human Activity Recognition
Leng, Z., Bhattacharjee, A., Rajasekhar, H., Zhang, L., Bruda, E., Kwon, H., & Plötz, T. (2024) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3678545 · Cited by 61
Self-citationPractically Adopting Human Activity Recognition
Xu, H., Zhou, P., Tan, R., & Li, M. (2023) · Proceedings of the 29th Annual International Conference on Mobile Computing and Networking · doi:10.1145/3570361.3613299 · Cited by 55
Cheng, D., Zhang, L., Bu, C., Wu, H., & Song, A. (2023) · Knowledge-Based Systems · doi:10.1016/j.knosys.2023.110789 · Cited by 55
Negative Selection by Clustering for Contrastive Learning in Human Activity Recognition
Wang, J., Zhu, T., Chen, L. L., Ning, H., & Wan, Y. (2023) · IEEE Internet of Things Journal · doi:10.1109/jiot.2023.3239945 · Cited by 52
“Some self-supervised learning works are applied to human activity recognition based on sensor data, such as Multitask SSL [33], CAE [38], Masked Reconstruction [39], and CPCHAR [40].”
MethodCross-Domain HAR: Few Shot Transfer Learning for Human Activity Recognition
Thukral, M., Haresamudram, H., & Ploetz, T. (2023) · arXiv · doi:10.48550/arxiv.2310.14390 · Cited by 45
Self-citationSanabria, A. R., Zambonelli, F., Dobson, S., & Ye, J. (2021) · Pervasive and Mobile Computing · doi:10.1016/j.pmcj.2021.101477 · Cited by 43
“Haresamudram et al. [22] have employed Contrastive Predictive Coding to learn temporal structures of sensor data.”
BackgroundApplication of human activity/action recognition: a review
sedaghati, N., Ardebili, S., & Ghaffari, A. (2025) · Multimedia Tools and Applications · doi:10.1007/s11042-024-20576-2 · Cited by 64
Cross-Domain HAR: Few-Shot Transfer Learning for Human Activity Recognition
Thukral, M., Haresamudram, H., & Plötz, T. (2024) · ACM Transactions on Intelligent Systems and Technology · doi:10.1145/3704921 · Cited by 70
“In [18] and [19], models pre-trained through self-supervised learning methods such as contrastive predictive coding, SimCLR and multi-task learning are tasked to recognize activities using supervision from a limited number of labeled windows per target activity class.”
InfluentialSelf-citationWu, Y., Lin, Q., Yang, M., Liu, J., Tian, J., Kapil, D., & Vanderbloemen, L. (2021) · Healthcare · doi:10.3390/healthcare10010036 · Cited by 59
“It has been successfully verified in many computer vision tasks such as image classification [17] and human activity recognition [18,19].”
BackgroundTransfer Learning in Sensor-Based Human Activity Recognition: A Survey
Dhekane, S. G., & Ploetz, T. (2025) · ACM Computing Surveys · doi:10.1145/3717608 · Cited by 42
“Unsupervised and self-supervised pre-training of neural networks addresses both the distribution shifts and the scarcity of annotations [78, 79, 180].”
Self-citationLipLearner: Customizable Silent Speech Interactions on Mobile Devices
Su, Z., Fang, S., & Rekimoto, J. (2023) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3544548.3581465 · Cited by 47
“HCI researchers have been applying FSL to tasks, such as sound recognition [29, 67] and human activity recognition [19], enabling in-situ model fine-tuning in the real world.”
MethodContrastive Accelerometer–Gyroscope Embedding Model for Human Activity Recognition
Koo, I., Park, Y., Jeong, M., & Kim, C. (2022) · IEEE Sensors Journal · doi:10.1109/jsen.2022.3222825 · Cited by 39
“The contrastive predictive coding (CPC) [38] infers the global structure of time-series data and predicts multiple subsequent timestamps.”
BackgroundAn active semi-supervised deep learning model for human activity recognition
Bi, H., Perelló-Nieto, M., Santos‐Rodríguez, R., Flach, P. A., & Craddock, I. J. (2022) · Journal of Ambient Intelligence and Humanized Computing · doi:10.1007/s12652-022-03768-2 · Cited by 39
“A contrastive predictive coding framework was developed (Haresamudram et al. 2021) to capture the underlying temporal structure of HAR data, leading to significantly improved recognition performance.”
MethodMiao, S., Chen, L., Hu, R., & Luo, Y. (2022) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3550331 · Cited by 36
“Tang et al. [46] and Haresamudram et al. [16] explored the self-supervised learning in WHAR.”
BackgroundKhaertdinov, B., Asteriadis, S., & Ghaleb, E. (2022) · IEEE Transactions on Biometrics Behavior and Identity Science · doi:10.1109/tbiom.2022.3180591 · Cited by 40
“Finally, as in other applications, contrastive SSL algorithms, namely contrastive predictive coding [9] and adapted SimCLR [10], demonstrate state-of-the-art performance for the sensor-based HAR task.”
BackgroundMethodInfluentialHaresamudram, H., Tang, C. I., Suh, S., Lukowicz, P., & Plötz, T. (2025) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3729467 · Cited by 33
“Alternatively, Hong et al. [84], improve cross dataset performance via masked modeling in conjunction with contrastive regularization.”
InfluentialSelf-citationLogacjov, A., & Bach, K. (2023) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2023.107478 · Cited by 41
Cheng, D., Zhang, L., Qin, L., Wang, S., Wu, H., & Song, A. (2024) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2024.3373019 · Cited by 31
“The primary objective of the study is to address one crucial issue present in existing HAR methods [2], [3], [6], [9], [10], [30], [32], which are dealing with annotation scarcity in sensor data by randomly masking out missing sensor readings and constructing an effective pre-training task, while…”
BackgroundMethodInfluentialKwon, H., Abowd, G. D., & Plötz, T. (2021) · Sensors · doi:10.3390/s21248337 · Cited by 30
“Recently, semi-supervised and especially self-supervised learning methods [101–103] have become popular in many machine learning application domains including human activity recognition using wearables [13,43,44].”
BackgroundSelf-citationContrastive Learning with Cross-Modal Knowledge Mining for Multimodal Human Activity Recognition
Brinzea, R., Khaertdinov, B., & Asteriadis, S. (2022) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn55064.2022.9892522 · Cited by 37
“In the field of HAR, contrastive SSL has mainly been applied on individual data sources, such as sensors [35], [36], skeleton data [34] or visual data [37].”
Taghanaki, S. R., Rainbow, M. J., & Etemad, A. (2023) · IEEE Transactions on Human-Machine Systems · doi:10.1109/thms.2023.3303438 · Cited by 31
“This approach has been used in prior works, such as [9], [47], and [13].”
MethodInvestigating Enhancements to Contrastive Predictive Coding for Human Activity Recognition
Haresamudram, H., Essa, I., & Plötz, T. (2023) · 2023 IEEE International Conference on Pervasive Computing and Communications (PerCom) · doi:10.1109/percom56429.2023.10099197 · Cited by 27
“We investigate enhancements to components of the current CPC setup in the field of sensor-based HAR [21].”
BackgroundMethodInfluentialSelf-citationSelf-supervised Learning for Accelerometer-based Human Activity Recognition: A Survey
Logacjov, A. (2024) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3699767 · Cited by 40
ActSonic: Recognizing Everyday Activities from Inaudible Acoustic Wave Around the Body
Mahmud, S., Parikh, V., Liang, Q., Li, K., Zhang, R., Ajit, A., Gunda, V., Agarwal, D., Guimbretière, F. V., & Zhang, C. (2024) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3699752 · Cited by 42
“In this section, we provide an overview of related work focused on IMU, camera, and acoustic sensing-based activity recognition, and position the contribution of ActSonic within this landscape.”
Chen, W., Cheng, J., Wang, L., Zhao, W., & Matusik, W. (2024) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3699747 · Cited by 48
“], including unsupervised domain adaptation [9], contrastive predictive coding [38], and deep unsupervised clustering [54].”
Spatial-Temporal Masked Autoencoder for Multi-Device Wearable Human Activity Recognition
Miao, S., Chen, L., & Hu, R. (2023) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3631415 · Cited by 23
“Leave-one-group-out cross-validation approach is used for performance evaluation follows the previous works [26, 34] .”
BackgroundMethodInfluentialSelfPAB: large-scale pre-training on accelerometer data for human activity recognition
Logacjov, A., Herland, S., Ustad, A., & Bach, K. (2024) · Applied Intelligence · doi:10.1007/s10489-024-05322-3 · Cited by 35
“In [14], the objective was to predict k future time steps starting from t inside a given time frame W i while the positive sample was the interval t to t + k in W i and the negative samples t to t + k in other windows W j (cid:2)= i .”
CroSSL: Cross-modal Self-Supervised Learning for Time-series through Latent Masking
Deldari, S., Spathis, D., Malekzadeh, M., Kawsar, F., Salim, F. D., & Mathur, A. (2024) · ACM International Conference on Web Search and Data Mining (WSDM) · doi:10.1145/3616855.3635795 · Cited by 56
Thukral, M., Dhekane, S. G., Hiremath, S. K., Haresamudram, H., & Ploetz, T. (2024) · arXiv · doi:10.48550/arxiv.2405.12368 · Cited by 21
Self-citationTemporal Feature Alignment in Contrastive Self-Supervised Learning for Human Activity Recognition
Khaertdinov, B., & Asteriadis, S. (2022) · IEEE International Joint Conference on Biometrics (IJCB) · doi:10.1109/ijcb54206.2022.10007984 · Cited by 28
“As can be seen from the table, the TFA module shows better performance than Contrastive Predictive Coding from [14] and the plain CSSHAR.”
BackgroundMethodBrand, Y. E., Kluge, F., Palmerini, L., Paraschiv-Ionescu, A., Becker, C., Cereatti, A., Maetzler, W., Sharrack, B., Vereijken, B., Yarnall, A. J., et al. (2024) · Scientific Reports · doi:10.1038/s41598-024-71491-3 · Cited by 19
Contrastive Self-Supervised Learning for Sensor-Based Human Activity Recognition: A Review
Chen, H., Gouin-Vallerand, C., Bouchard, K., Gaboury, S., Couture, M., Bier, N., & Giroux, S. (2024) · IEEE Access · doi:10.1109/access.2024.3480814 · Cited by 29
“CPCHAR [39] adopts the CPC framework to learn effective sensor representations in an unsupervised manner by leveraging the temporal characteristics of body-worn sensor data.”
MethodThukral, M., Dhekane, S. G., Hiremath, S. K., Haresamudram, H., & Plötz, T. (2025) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3712278 · Cited by 51
Self-citationRay, S. J., Cherian, J., Liberty, A. M., Hammond, T. A., & Shireman, P. K. (2024) · doi:10.2196/preprints.67373 · Cited by 18
Motifs-based recommender system via hypergraph convolution and contrastive learning
Sun, Y., Zhu, D., Du, H., & Tian, Z. (2022) · Neurocomputing · doi:10.1016/j.neucom.2022.09.102 · Cited by 24
“Specifically, in each calculation process, one positive sample (𝑘+) and 𝐾 negative samples (𝑘0, ..., 𝑘𝐾 ) from the queue are selected for calculation, and InfoNCE[10] can be used to perform the following calculation: ℓ = − log exp(ℎ⊤𝑞 ℎ𝑚𝑘+ /𝜏)∑𝐾 𝑖=0 exp(ℎ⊤𝑞 ℎ𝑚𝑘𝑖 /𝜏) (10) where ℎ𝑞 is the representation of the query nod…”
MethodInfluentialSun, Y., Xu, X., Tian, X., Zhou, L., & Li, Y. (2024) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2024.108705 · Cited by 26
Ray, S. J., Cherian, J., Liberty, A. M., Hammond, T. A., & Shireman, P. K. (2025) · Journal of Medical Internet Research · doi:10.2196/67373 · Cited by 18
If only we had more data!: Sensor-Based Human Activity Recognition in Challenging Scenarios
Plötz, T. (2023) · 2023 IEEE International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events (PerCom Workshops) · doi:10.1109/percomworkshops56833.2023.10150267 · Cited by 16
“In [50] (and refined in [51]) we introduced the Contrastive Predictive Coding (CPC) framework to human activity recognition, which captures the temporal structure Fig.”
MethodSelf-citationSelf-Supervised WiFi-Based Activity Recognition
Bocus, M. J., Lau, H.-S., McConville, R., Piechocki, R. J., & Santos‐Rodríguez, R. (2022) · IEEE Globecom Workshops (GC Wkshps) · doi:10.1109/gcwkshps56602.2022.10008537 · Cited by 29
“In the same context, the authors of [13] investigated the effectiveness of Contrastive Predictive Coding (CPC) framework for HAR when applied to various wearable sensor datasets.”
BackgroundSun, K., Xia, C., Zhang, X., Chen, H., & Zhang, J. (2024) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3643553 · Cited by 21
“SSL has been introduced recently to train feature representation models based on motion sensor data [24, 36–38].”
BackgroundMethodMESEN: Exploit Multimodal Data to Design Unimodal Human Activity Recognition with Few Labels
Xu, L., Gu, C., Tan, R., He, S., & Chen, J. (2023) · ACM Conference on Embedded Networked Sensor Systems (SenSys) · doi:10.1145/3625687.3625782 · Cited by 26
Haresamudram, H., Essa, I., & Plötz, T. (2024) · Sensors · doi:10.3390/s24041238 · Cited by 14
“Contrastive Predictive Coding (CPC) was adopted and applied to wearable sensor data in [68], where future timestep prediction was performed under contrastive learning settings.”
BackgroundMethodInfluentialSelf-citationRobust Finger Interactions with COTS Smartwatches via Unsupervised Siamese Adaptation
Chen, W., Wang, Z., Quan, P., Peng, Z., Lin, S., Srivastava, M. B., Matusik, W., & Stankovic, J. A. (2023) · ACM Symposium on User Interface Software and Technology (UIST) · doi:10.1145/3586183.3606794 · Cited by 20
“Besides, most of the domain adversarial learning solutions have only been used in image classification [24, 57] or large-scale movements such as coarse-grained human activities [7, 29].”
Background1D Convolutional Neural Network with Long Short-Term Memory for Human Activity Recognition
Goh, J. X., Lim, K. M., & Lee, C. P. (2021) · IEEE International Conference on Artificial Intelligence in Engineering and Technology (IICAIET) · doi:10.1109/iicaiet51634.2021.9573979 · Cited by 14
Chen, X., Zhou, X., Sun, M., & Wang, H. (2025) · IEEE Sensors Journal · doi:10.1109/jsen.2024.3491933 · Cited by 13
“Comparing supervised TCLHAR with TCLHAR and supervised CSSHAR with CSSHAR, we can see that self-supervised models always perform worse than the supervised learning models, which can also be seen in existing studies [8], [9], [10].”
InfluentialSoK: Behind the Accuracy of Complex Human Activity Recognition Using Deep Learning
Nguyen, D., & Le‐Khac, N. (2024) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn60899.2024.10650322 · Cited by 19
“[53]2021 Data (variability) Pre-training on unlabelled data using the Contrastive Predictive Coding (CPC) framework, fine-tuning on labelled data for classification.”
MethodGOAT: A Generalized Cross-Dataset Activity Recognition Framework with Natural Language Supervision
Miao, S., & Chen, L. (2024) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3699736 · Cited by 27
“Self-supervised methods [13, 30, 47, 61] have emerged as a powerful paradigm to exploit large-scale unlabeled data.”
DiversityOne: A Multi-Country Smartphone Sensor Dataset for Everyday Life Behavior Modeling
Busso, M., Bontempelli, A., Malcotti, L. J., Meegahapola, L., Kun, P., Diwakar, S., Nutakki, C. S., Britez, M. D. R., Xu, H., Song, D., et al. (2025) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3712289 · Cited by 26
MassNet: A Deep Learning Approach for Body Weight Extraction from A Single Pressure Image
Wu, Z., Wan, Q., Zhao, M., Ke, Y., Fang, Y., Liang, Z., Xie, F., & Cheng, J. (2023) · IEEE International Conference on Pervasive Computing and Communications · doi:10.1109/percom56429.2023.10099301 · Cited by 12
“Contrastive learning has achieved noticeable performance in many downstream tasks such as human activity recognition and biosignal process [11], [25].”
BackgroundContrastSense: Domain-invariant Contrastive Learning for In-the-Wild Wearable Sensing
Dai, G., Xu, H., Yoon, H., Li, M., Tan, R., & Lee, S.-J. (2024) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3699744 · Cited by 27
“CPCHAR [16] conducts contrastive pretraining on the unlabeled data by predicting the nearby features.”
InfluentialClustering of Human Activities from Wearables by Adopting Nearest Neighbors
Ahmed, A., Haresamudram, H., & Plötz, T. (2022) · ACM International Symposium on Wearable Computers (ISWC) · doi:10.1145/3544794.3558477 · Cited by 14
“Using the protocol detailed in [19], we also compare the performance of our adaptation of SimCLR against overlapping datasets present in the state-of-the-art, CPC, which obtains 80.05% and 89.36% for Mobiact and Motionsense respectively, relative to our 89.64% and 89.54%.”
BackgroundMethodInfluentialSelf-citationWang, J., Cui, W., Zhu, T., Ning, H., & Liu, Z. (2024) · IEEE Sensors Journal · doi:10.1109/jsen.2024.3390755 · Cited by 13
“The study [32] was the first to apply CPC [33] to the pretraining of activity data, achieving superior performance compared to previous approaches.”
MethodComparing self-supervised learning techniques for wearable human activity recognition
Ek, S., Presotto, R., Civitarese, G., Portet, F., Lalanda, P., & Bettini, C. (2025) · CCF Transactions on Pervasive Computing and Interaction · doi:10.1007/s42486-024-00182-9 · Cited by 15
“Other studies proposed generative meth-ods (Varamin et al. 2018; Haresamudram et al. 2020; Zhang et al. 2022) and demonstrated competitive performances.”
How Much Unlabeled Data is Really Needed for Effective Self-Supervised Human Activity Recognition?
Dhekane, S. G., Haresamudram, H., Thukral, M., & Plötz, T. (2023) · Proceedings of the 2023 International Symposium on Wearable Computers · doi:10.1145/3594738.3611366 · Cited by 9
“Contrastive Predictive Coding, which uses future timestep prediction in a contrastive setup [18, 25].”
InfluentialSelf-citationTFC: Time–frequency contrasting network for wearable-based human activity recognition
Huang, Z., Deng, J., Wang, S., Tang, C., & Xiao, S. (2025) · Knowledge-Based Systems · doi:10.1016/j.knosys.2025.113373 · Cited by 9
Human Activity Recognition via Temporal Fusion Contrastive Learning
Kim, I., Lim, J., & Lee, J. (2024) · IEEE Access · doi:10.1109/access.2024.3357143 · Cited by 10
“Haresamudram et al. [13] applied the CPC [14] framework to HAR.”
MethodInfluentialThe Human Continuity Activity Semisupervised Recognizing Model for Multiview IoT Network
Yuan, R., & Wang, J. P. (2023) · IEEE Internet of Things Journal · doi:10.1109/jiot.2023.3234053 · Cited by 8
“To reduce the reliance on manual labels, many researchers have introduced semisupervised activity feature extraction into HAR [15], [17], [18].”
BackgroundMethodInfluentialMachine learning for medical image classification
Husain, G., Mayer, J., Bekbolatova, M., Vathappallil, P., Matalia, M., & Toma, M. (2024) · Academia Medicine · doi:10.20935/acadmed7444 · Cited by 15
Enhancing Sensor-Based Human Activity Recognition With Multiresolution Wavelet-Attention
Wieland, C., & Pankratius, V. (2025) · IEEE Sensors Journal · doi:10.1109/jsen.2025.3571572 · Cited by 14
“Common pretraining objectives include signal reconstruction [36], masked signal modeling [37], signal forecasting [38], transformation detection [39], and contrastive learning [40].”
MobHAR: Source-free Knowledge Transfer for Human Activity Recognition on Mobile Devices
Xue, M., Zhu, Y., Xie, W., Wang, Z., Chen, Y., Jiang, K., & Zhang, Q. (2025) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3712620 · Cited by 14
“Existing HAR models such as [17, 19, 21, 27, 40, 45, 46, 56, 57] commonly assume that data adheres to the concept of independent and identically distributed (IID) [50].”
WaveHAR: Learning Wavelet Representation for Wearable Human Activity Recognition
Yao, M., Zhang, L., Cheng, D., Wu, H., & Song, A. (2025) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2025.3571487 · Cited by 13
“Haresamudram et al. [23] introduced a novel contrastive predictive coding technique that can maintain good recognition performance even when the number of labeled activity samples is extremely low.”
An, S., Gazi, A. H., & Inan, O. T. (2022) · IEEE Sensors Journal · doi:10.1109/jsen.2022.3194677 · Cited by 8
“representation learning of body-worn sensor data [28].”
MethodChugh, G., Chakraborty, S., Bhandari, R. B., & Chakraborty, S. (2023) · IEEE/ACM Conference on Connected Health: Applications, Systems and Engineering Technologies · doi:10.1145/3580252.3586981 · Cited by 9
“Recently, researchers [13, 24, 32, 34, 39, 42] studied the possibility of determining multi-task activities from sensory data.”
BackgroundChen, Y., Hu, C., Kimura, T., Li, Q., Liu, S., Wu, F., & Chen, G. (2024) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3699779 · Cited by 9
“…pairs (with proxy label 1) against semantically separate negative pairs (with proxy label 0), where the positive pairs could be defined based on transformation invariance [3, 12, 39], multi-view synchronization [41, 45], temporal locality [46, 55], or local-to-global correspondence [8, 27].”
Guo, P., & Nakayama, M. (2025) · Sensors · doi:10.3390/s25226988 · Cited by 6
“Other methods, such as CPC [30] and ClusterCLHAR [35], also involved two-stage training.”
Wu, Y., Huang, W., & Zhao, Y. (2025) · Journal of Energy Storage · doi:10.1016/j.est.2025.115955 · Cited by 9
Self-Supervised Approach for Few-shot Hand Gesture Recognition
Kimura, N. (2022) · ACM Symposium on User Interface Software and Technology (UIST) · doi:10.1145/3526114.3558707 · Cited by 6
“For a task with a well-developed dataset (human-activity detection), Haresamudram et al. used contrastive predictive coding [5].”
MethodWearable Sensor-Based Human Activity Recognition for Worker Safety in Manufacturing Line
Suh, S., Rey, V. F., & Lukowicz, P. (2023) · Springer eBooks · doi:10.1007/978-3-031-46452-2_17 · Cited by 15
Activity Recognition from Channel State Information for Few-Sampled Scenarios
(2023) · doi:10.1109/gcwkshps58843.2023.10465069 · Cited by 5
“Concerning HAR, in [10] data from wearable sensors are used to obtain long-term temporal structure and apply fine-tuning to small sets of labeled training data.”
MethodDon't freeze: Finetune encoders for better Self-Supervised HAR
Rey, V. F., Nshimyimana, D., & Lukowicz, P. (2023) · ACM International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp) · doi:10.1145/3594739.3610790 · Cited by 10
“cpc[4, 6] uses an encoder that generates a representation for each timestep.”
MethodSelf-Supervised Federated Learning for Personalized Human Activity Recognition
Deng, S., Teng, D., Guo, Z., Chen, J., Chen, D., Jia, T., & Wang, H. (2024) · IEEE International Conference on Multimedia and Expo (ICME) · doi:10.1109/icme57554.2024.10687970 · Cited by 6
“The contrastive predictive code is introduced to train models by comparing predictive coding in proxy tasks [14].”
MethodKhan, M. N. H., Creswell, J. D., Albert, J., O'Connell, P., Fallon, S., Polowitz, M., Xu, X., & Islam, B. (2025) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3749498 · Cited by 6
“We use raw accelerometer data rather than the filtered or processed data from the previous section, as the deep learning model is capable of denoising the data itself and capturing relevant information [36, 56].”
Self-Supervised Learning for Activity Recognition Based on Datasets With Imbalanced Classes
Hamad, R. A., Woo, W. L., Wei, B. W., & Yang, L. (2025) · IEEE Transactions on Emerging Topics in Computational Intelligence · doi:10.1109/tetci.2025.3526504 · Cited by 11
“Haresamudram, et al. [31] proposed a self-supervised method based on contrastive predictive coding (CPC).”
Semi-supervised Human Activity Recognition with individual difference alignment
Yang, Z., Li, T., Xu, Z., Huang, Z., Cao, Y., Li, K., & Ma, J. (2025) · Expert Systems with Applications · doi:10.1016/j.eswa.2025.126976 · Cited by 4
OpenMAE: Efficient Masked Autoencoder for Vibration Sensing with Open-domain Data Enrichment
Hu, C., Chen, Y., Kara, D., Liu, S., Abdelzaher, T., Wu, F., & Chen, G. (2025) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3729485 · Cited by 6
Long-term self-supervised learning for accelerometer-based sleep–wake recognition
Logacjov, A., Bach, K., & Mork, P. J. (2024) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2024.109758 · Cited by 8
MultiHGR: Multi-Task Hand Gesture Recognition with Cross-Modal Wrist-Worn Devices
Lyu, M., Zhou, H., Guo, K., Zhou, W., Shen, X., & Gu, Y. (2024) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom52122.2024.10621430 · Cited by 4
“Haresamudram et al. [25] leveraged the Contrastive Predictive Coding (CPC) framework and the inherent temporal structure of sensor data streams to achieve self-supervised learning.”
BackgroundRey, V. F., Rezende, P. M. B., Zhou, B., Suh, S., & Lukowicz, P. (2025) · Expert Systems with Applications · doi:10.1016/j.eswa.2025.129288 · Cited by 14
Khaertdinov, B., & Asteriadis, S. (2023) · IEEE International Joint Conference on Biometrics (IJCB) · doi:10.1109/ijcb57857.2023.10448965 · Cited by 5
“While earlier works include transformation-based [26] and reconstruction approaches [14], the latter works are based on contrastive learning frameworks, such as Contrastive Predictive Coding (CPC) and SimCLR [15, 19, 17], where models are trained to pull together augmented views of input data.”
BackgroundMethodA Deep Learning Framework for Classifying and Evaluating Yoga Exercises
Malik, M. L., & Khosla, A. (2023) · Journal of Computer Science · doi:10.3844/jcssp.2023.229.241 · Cited by 3
Self-Supervised Learning for Complex Activity Recognition Through Motif Identification Learning
Xia, Q., Morales, J., Huang, Y., Hara, T., Wu, K., Oshima, H., Fukuda, M., Namioka, Y., & Maekawa, T. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3514736 · Cited by 7
“CPC [47]: CPC employs an LSTM network designed to predict subsequent time-series segments.”
Human Activity Classification Using Recurrence Plot and Residual Network
Lew, C. H., Lim, K. M., Lee, C. P., & Lim, J. Y. (2023) · IEEE Conference on Systems, Process and Control · doi:10.1109/icspc59664.2023.10420336 · Cited by 3
Yang, Z., Li, K., Huang, Z., Xu, Z., Zhu, X., & Xiao, Y. (2025) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2025.112317 · Cited by 3
Segment-Based Unsupervised Learning Method in Sensor-Based Human Activity Recognition
Takenaka, K., Kondo, K., & Hasegawa, T. (2023) · Sensors · doi:10.3390/s23208449 · Cited by 5
“Focusing on time series, Haresamudram et al. [12] proposed a method to use contrastive predictive coding [37] for HAR.”
BackgroundMethodDeepSense++: Robust HAR with Missing Data
gowdaman, S. K., & Akther, S. (2026) · IEEE/SICE International Symposium on System Integration (SII) · doi:10.1109/sii64115.2026.11404464 · Cited by 3
“Haresamudram et al. [13] applied contrastive predictive coding for downstream classification, and Song et al. [14] explored contrastive pretraining for ambient sensing.”
Zhang, D., Wong, W. K., & Chew, I. M. (2025) · International Journal of Multimedia Information Retrieval · doi:10.1007/s13735-025-00382-8 · Cited by 5
Metadata
- Authors
- Haresamudram, H., Essa, I., & Plötz, T.
- Affiliations
- Georgia Institute of Technology (US)
- Venue
- Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Published
- 2021-06-23
- Volume
- 5
- Issue
- 2
- Pages
- 1-26
- ISSN
- 2474-9567 (ISSN-L)
- OpenAlex
- W3111841649
- Cited by
- 150
- References
- 99
- Citations by year
- 2026: 15 · 2025: 42 · 2024: 39 · 2023: 27 · 2022: 21 · 2021: 6
- FWCI
- 8.35
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies (publishedVersion, public-domain)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- 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
- Social Sciences › Transportation › Human Mobility and Location-Based Analysis
- Keywords
- human activity recognition, contrastive predictive coding, self-supervised learning, wearable sensors, motion sensors, learned representations, pre-training, limited labeled data, wearable computing
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
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
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.