Ensembles of Deep LSTM Learners for Activity Recognition using Wearables
Open accessThe DOI matches its Crossref record; 53 of the 65 references it cites were found in a registry or an index, and 10 are not checked yet; none of the 62 sources its claims cite is checked for support yet.
Abstract
Recently, deep learning (DL) methods have been introduced very successfully into human activity recognition (HAR) scenarios in ubiquitous and wearable computing. Especially the prospect of overcoming the need for manual feature design combined with superior classification capabilities render deep neural networks very attractive for real-life HAR applications. Even though DL-based approaches now outperform the state-of-the-art in a number of recognition tasks, still substantial challenges remain. Most prominently, issues with real-life datasets, typically including imbalanced datasets and problematic data quality, still limit the effectiveness of activity recognition using wearables. In this paper we tackle such challenges through Ensembles of deep Long Short Term Memory (LSTM) networks. LSTM networks currently represent the state-of-the-art with superior classification performance on relevant HAR benchmark datasets. We have developed modified training procedures for LSTM networks and combine sets of diverse LSTM learners into classifier collectives. We demonstrate that Ensembles of deep LSTM learners outperform individual LSTM networks and thus push the state-of-the-art in human activity recognition using wearables. Through an extensive experimental evaluation on three standard benchmarks (Opportunity, PAMAP2, Skoda) we demonstrate the excellent recognition capabilities of our approach and its potential for real-life applications of human activity recognition.
Checked 2026-10-05 · 10 references not checked yet
Overview
CoChat Summary
TL;DR
The paper introduces ensembles of deep LSTM learners for wearable activity recognition, using an epoch-wise bagging-style training procedure and score-level fusion of multiple LSTM models. Across Opportunity, PAMAP2, and Skoda, the ensembles outperform individual LSTMs and state-of-the-art baselines, with the largest gains on the challenging, imbalanced Opportunity dataset.
Key findings
- Ensembles of LSTM learners improve over single LSTMs on all three benchmarks. For Opportunity, single CE LSTM F1 = 0.674 ± 0.021 vs. Ensemble M=10 CE = 0.716 ± 0.011 and M=20 CE = 0.720 ± 0.011; the best combined CE+F1 ensemble reaches 0.726 ± 0.008. For PAMAP2, single CE = 0.802 ± 0.045 vs. Ensemble M=10 CE = 0.848 ± 0.021 and M=20 CE = 0.852 ± 0.026; combined CE+F1 = 0.854 ± 0.026. For Skoda, single CE = 0.918 ± 0.007 vs. Ensemble M=10 CE = 0.925 ± 0.002 and M=20 CE = 0.926 ± 0.001; combined CE+F1 = 0.924 ± 0.002.
- The LSTM ensemble outperforms reported state-of-the-art baselines: LSTM baseline 0.659/0.756/0.904, DeepConvLSTM 0.672/0.748/0.912, and LSTM-S 0.684/0.838/0.921 for Opportunity/PAMAP2/Skoda, respectively, versus the ensemble’s 0.726/0.854/0.924.
- Performance generally plateaus at M=10 base learners; increasing to M=20 gives no statistically significant improvement except for Opportunity with F1 loss, where the improvement is “just significant.” Combining CE and F1 loss functions further improves Opportunity but not PAMAP2, and Skoda shows no substantial boost.
- The training procedure intentionally omits data: on average about 35%–38% of training data is not used per epoch when using a uniformly distributed random number generator for training frames.
Methods
- The authors use two-layer LSTM networks with 256 LSTM units per layer, dropout 0.5 at both hidden layers, ADAM with learning rate 0.001, implemented in Lasagne/Theano and trained on an NVIDIA Tesla K40. Training uses randomly drawn mini-batch sizes from U(128, 256) and frame lengths from U(16, 32); test-time inference is sample-wise with mini-batch size and frame length set to 1.
- The core method is epoch-wise bagging: for each epoch, random mini-batch size, random starting positions, and random frame lengths are drawn; one LSTM learner is produced per epoch (MaxEpoch = 100). The M best models on a validation set are retained and fused by arithmetic mean of their softmax class-probability vectors. Both cross-entropy and F1 loss are explored.
- Evaluation uses three standard benchmarks: Opportunity, PAMAP2, and Skoda, with hold-out validation/test protocols replicating prior work. Results are mean F1 scores over 30 repetitions, with two-tailed independent t-tests for significance.
Limitations
- The paper states that implementation issues for field deployments or products are not yet addressed; ensembles inflate the parameter space, though the authors argue pruning remains accessible because the base LSTM models are unchanged.
- The paper notes that F1 loss is a global measure and cannot directly measure individual sample losses; because LSTMs are trained on mini-batches, it may not precisely approximate the true gradient toward optimal F1 over the whole training set.
- The paper states that Skoda shows no substantial performance boost, attributed to only one worker being considered and the task lacking challenging test data from different environments; the baseline models already achieve high specialisation (F1 > 0.9). Evident from the text: evaluation is limited to three benchmark datasets and does not include an in-the-wild deployment, and frame-based results from prior work were converted to sample-wise predictions for comparability.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
0 Valid · 11 Not valid · 48 Not checked · 80 Own results and interpretations
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Through an extensive experimental evaluation on three standard benchmarks (Opportunity, PAMAP2, Skoda) we demonstrate the excellent recognition capabilities of our approach and its potential for real-life applications of human activity recognition.”
boosted- 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.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“By outperforming the state-of-the-art our results indicate increased robustness of our Ensembles of deep LSTM learners with regards to typical challenges of human activity recognition in ubiquitous and wearable computing scenarios.”
hedgedstates 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.
- Rests on an assumption not shown here.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We demonstrate that Ensembles of deep LSTM learners outperform individual LSTM networks and thus push the state-of-the-art in human activity recognition using wearables.”
boostedstates 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.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“strong learners -LSTM models in our case- can indeed be integrated into very effective ensembles that outperform single, that is, individual models. Recognition results are consistent across all three tasks”
boostedstates 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.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“PAMAP2 does not benefit from diversity in loss function”
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.
55 more claims
Create a free account to see all 139 claims and to verify the 62 references still open.
- It may state two separate claims as one. 32
- The quote only partly states this claim. 2
- Probably not a substantial claim. 59
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 says what the claim attributes to it.
In the citing paper
“Many wearables datasets exhibit substantial class imbalance [68].”
From Past, Present, and Future of Sensor-based Human Activity Recognition Using Wearables: A Surveying Tutorial on a Still Challenging Task (2025) · cites this paper as [68]
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
§ 3.2 Modified Training for LSTM Models · p. 12
“Overall, only a total of ˆ B ⌊ T/ ˆ B ⌋ samples are used in each epoch.”
§ ALGORITHM 1: Epoch-wise bagging for LSTM learners. · p. 13
“while L seq ≤ ⌊ T/ ˆ B j ⌋ do draw random integer ˆ L i (i.e., frame length) from discrete uniform distribution U ( L low , L high ) ;”
Citations
Cites
Breiman, L. (2001) · Machine Learning · doi:10.1023/a:1010933404324 · Cited by 133,538
Hochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318
Adam: 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
The Elements of Statistical Learning
Hastie, T., Tibshirani, R., & Friedman, J. (2009) · Springer Series in Statistics · doi:10.1007/978-0-387-84858-7 · Cited by 25,970
The Elements of Statistical Learning
Hastie, T., Friedman, J., & Tibshirani, R. (2001) · Springer Series in Statistics · doi:10.1007/978-0-387-21606-5 · Cited by 24,162
Breiman, L. (1996) · Machine Learning · doi:10.1007/bf00058655 · Cited by 16,325
Parallel Distributed Processing: Explorations in the Microstructure of Cognition: Foundations
Rumelhart, D. E., McClelland, J. L., & AU (1986) · doi:10.7551/mitpress/5236.001.0001 · Cited by 15,686
The Computer for the 21st Century
Weiser, M. (1991) · Scientific American · doi:10.1038/scientificamerican0991-94 · Cited by 10,017
Large-Scale Machine Learning with Stochastic Gradient Descent
Bottou, L. (2010) · Proceedings of COMPSTAT'2010 · doi:10.1007/978-3-7908-2604-3_16 · Cited by 5,795
Research works printed but not linked: 2
- James L McClelland and David E Rumelhart . 1986. Parallel distributed processing . MIT Press . James L McClelland and David E Rumelhart. 1986. Parallel distributed processing. MIT Press.
- Theano Development Team . 2016 . Theano: A Python framework for fast computation of mathematical expressions. arXiv e-prints abs/1605.02688 (May 2016). http://arxiv.org/abs/1605.02688 Theano Development Team. 2016. Theano: A Python framework for fast computation of mathematical expressions. arXiv e-prints abs/1605.02688 (May 2016). http://arxiv.org/abs/1605.02688 · arXiv
Cited by
Deep learning for sensor-based activity recognition: A survey
Wang, J., Chen, Y., Hao, S., Peng, X., & Hu, L. (2019) · Pattern Recognition Letters · doi:10.1016/j.patrec.2018.02.010 · Cited by 2,014
“(Guan and Ploetz, 2017) Body-worn, object, ambient RNN ADL, smart home D01, D02, D04”
BackgroundMethodNweke, H. F., Teh, Y. W., Al-garadi, M. A., & Alo, U. R. (2018) · Expert Systems with Applications · doi:10.1016/j.eswa.2018.03.056 · Cited by 930
Deep Learning for Sensor-based Human Activity Recognition
Chen, K., Zhang, D., Yao, L., Guo, B., Yu, Z., & Liu, Y. (2022) · ACM Computing Surveys · doi:10.1145/3447744 · Cited by 690
“st a certain percentage of data in the training set. Another direction of solving the imbalance concern is to modify the model-building strategy instead of directly balancing the training dataset. In [49], Guan and Plötz utilized the F1-score rather than the conventional cross-entropy as the loss function to address t…”
BackgroundMethodQiu, S., Zhao, H., Jiang, N., Wang, Z., Liu, L., An, Y., Zhao, H., Miao, X., Liu, R., & Fortino, G. (2022) · Information Fusion · doi:10.1016/j.inffus.2021.11.006 · Cited by 671
Zero-Effort Cross-Domain Gesture Recognition with Wi-Fi
Zheng, Y., Zhang, Y., Qian, K., Zhang, G., Liu, Y., Wu, C., & Yang, Z. (2019) · Proceedings of the 17th Annual International Conference on Mobile Systems, Applications, and Services · doi:10.1145/3307334.3326081 · Cited by 557
“Traditional approaches use cameras [16, 24, 42], wearable devices and phones [8, 17, 36] or sonar [22, 29, 48] as the sensing module.”
MethodDeep PPG: Large-Scale Heart Rate Estimation with Convolutional Neural Networks
Reiss, A., Indlekofer, I., Schmidt, P., & Van Laerhoven, K. (2019) · Sensors · doi:10.3390/s19143079 · Cited by 377
“Therefore, since capturing more data variability during training, combining these models will lead to increased modelling capabilities [49].”
BackgroundApplication of artificial intelligence in wearable devices: Opportunities and challenges
Nahavandi, D., Alizadehsani, R., Khosravi, A., & Acharya, U. R. (2022) · Computer Methods and Programs in Biomedicine · doi:10.1016/j.cmpb.2021.106541 · Cited by 372
Nweke, H. F., Teh, Y. W., Mujtaba, G., & Al-garadi, M. A. (2019) · Information Fusion · doi:10.1016/j.inffus.2018.06.002 · Cited by 394
Demrozi, F., Pravadelli, G., Bihorac, A., & Rashidi, P. (2020) · IEEE Access · doi:10.1109/access.2020.3037715 · Cited by 368
“The second most used model was the Long Short-TermMemory (LSTM) model, which was used in 17 papers [7], [83], [89], [102], [107], [112], [125], [130], [137], [139], [152], [153], [172],”
MethodA Semisupervised Recurrent Convolutional Attention Model for Human Activity Recognition
Chen, K., Yao, L., Zhang, D., Wang, X., Chang, X., & Nie, F. (2020) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2019.2927224 · Cited by 368
“Although some approaches [10], [16] have been proposed to solve the imbalance issue, their performance deteriorates due to the small data size in semisupervised learning.”
BackgroundMethodWidar3.0: Zero-Effort Cross-Domain Gesture Recognition with Wi-Fi
Zhang, Y., Zheng, Y., Qian, K., Zhang, G., Liu, Y., Wu, C., & Yang, Z. (2021) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2021.3105387 · Cited by 312
“Traditional approaches use cameras [1], [2], [3], wearable devices and phones [4], [5], [6] or sonar [7], [8], [9] as the sensing module.”
A multibranch CNN-BiLSTM model for human activity recognition using wearable sensor data
Challa, S. K., Kumar, A., & Semwal, V. B. (2022) · The Visual Computer · doi:10.1007/s00371-021-02283-3 · Cited by 338
Deep Learning for Time Series Classification and Extrinsic Regression: A Current Survey
Mohammadi Foumani, N., Miller, L., Tan, C. W., Webb, G. I., Forestier, G., & Salehi, M. (2024) · ACM Computing Surveys · doi:10.1145/3649448 · Cited by 268
“…[202] 2015 CNN DeepConvLSTM [192] 2016 CNN, LSTM Hammerla et al. [199] 2016 CNN, LSTM Bi-directional Ronao and Cho [203] 2016 CNN Guan and Plötz [204] 2017 LSTM Ensemble Lee et al. [205] 2017 CNN Murad and Pyun [206] 2017 LSTM Uni-and bi-directional Ignatov [207] 2018 CNN Statistical features…”
BackgroundMethodDeep Learning for Sensor-based Human Activity Recognition: Overview, Challenges and Opportunities
Chen, K., Zhang, D., Yao, L., Guo, B., Yu, Z., & Liu, Y. (2020) · arXiv · doi:10.48550/arxiv.2001.07416 · Cited by 314
“st a certain percentage of data in the training set. Another direction of solving the imbalance concern is to modify the model-building strategy instead of directly balancing the training dataset. In [49], Guan and Plötz utilized the F1-score rather than the conventional cross-entropy as the loss function to address t…”
BackgroundMethodA Survey on Deep Learning for Human Activity Recognition
Gu, F., Chung, M.-H., Chignell, M., Valaee, S., Zhou, B., & Liu, X. (2022) · ACM Computing Surveys · doi:10.1145/3472290 · Cited by 272
“Murad and Pyun [143] Guan and Plötz [73] developed a HAR method that ensembles multiple LSTMs, and showed that the ensembles of LSTM networks outperform individual LSTM networks.”
BackgroundMethodEdge Intelligence: Empowering Intelligence to the Edge of Network
Xu, D., Li, T., Li, Y., Su, X., Tarkoma, S., Jiang, T., Crowcroft, J., & Hui, P. (2021) · Proceedings of the IEEE · doi:10.1109/jproc.2021.3119950 · Cited by 272
Daily Activity Feature Selection in Smart Homes Based on Pearson Correlation Coefficient
Liu, Y., Mu, Y., Chen, K., Li, Y., & Guo, J. (2020) · Neural Processing Letters · doi:10.1007/s11063-019-10185-8 · Cited by 283
Deep CNN-LSTM With Self-Attention Model for Human Activity Recognition Using Wearable Sensor
Khatun, M. A., Yousuf, M. A., Ahmed, S., Uddin, M. Z., Alyami, S. A., Al-Ashhab, S., Akhdar, H. F., Khan, A., Azad, A., & Moni, M. A. (2022) · IEEE Journal of Translational Engineering in Health and Medicine · doi:10.1109/jtehm.2022.3177710 · Cited by 288
Human Activity Recognition Using Inertial Sensors in a Smartphone: An Overview
Sousa Lima, W., Souto, E., El-Khatib, K., Jalali, R., & Gama, J. (2019) · Sensors · doi:10.3390/s19143213 · Cited by 256
AttnSense: Multi-level Attention Mechanism For Multimodal Human Activity Recognition
Ma, H., Li, W., Zhang, X., Gao, S., & Lu, S. (2019) · Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2019/431 · Cited by 266
Leveraging Machine Learning and Big Data for Smart Buildings: A Comprehensive Survey
Qolomany, B., Al-Fuqaha, A., Gupta, A., Benhaddou, D., Alwajidi, S., Qadir, J., & Fong, A. C. (2019) · IEEE Access · doi:10.1109/access.2019.2926642 · Cited by 246
“Guan and Ploetz [124] proposed a deep LSTM ensemble”
BackgroundDeep learning for edge computing applications: A State-of-the-Art survey
Wang, F., Zhang, M., Wang, X., Ma, X., & Liu, J. (2020) · IEEE Access · doi:10.1109/access.2020.2982411 · Cited by 259
“RNN and LSTM are widely used in various sequential scenarios, such as language processing [43] and activity recognition [44].”
BackgroundKumar, P., Chauhan, S., & Awasthi, L. K. (2024) · Archives of Computational Methods in Engineering · doi:10.1007/s11831-023-09986-x · Cited by 206
“The authors of [151] used a HAR method to show that LSTM networks working together produce better results than those working alone.”
Multiscale Deep Feature Learning for Human Activity Recognition Using Wearable Sensors
Tang, Y., Zhang, L., Min, F., & He, J. (2023) · IEEE Transactions on Industrial Electronics · doi:10.1109/tie.2022.3161812 · Cited by 224
“Guan et al. [17] ensembled multiple LSTM units, which outperforms LSTM networks alone and significantly pushes the state-of-the-art in activity recognition tasks.”
BackgroundEdge Intelligence: Architectures, Challenges, and Applications
Xu, D., Li, T., Li, Y., Su, X., Tarkoma, S., Jiang, T., Crowcroft, J., & Hui, P. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2003.12172 · Cited by 120
“, healthcare, fitness tracking, and activity monitoring [379], [380].”
BackgroundLonini, L., Dai, A., Shawen, N., Simuni, T., Poon, C., Shimanovich, L., Daeschler, M., Ghaffari, R., Rogers, J. A., & Jayaraman, A. (2018) · npj Digital Medicine · doi:10.1038/s41746-018-0071-z · Cited by 209
Zeng, M., Gao, H., Yu, T., Mengshoel, O. J., Langseth, H., Lane, I., & Liu, X. (2018) · Proceedings of the 2018 ACM International Symposium on Wearable Computers · doi:10.1145/3267242.3267286 · Cited by 185
“[8] used ensemble LSTM model to capture diverse data during the training and significantly improved the recognition performance and robustness.”
BackgroundMethodResultInfluentialElf: accelerate high-resolution mobile deep vision with content-aware parallel offloading
Zhang, W., He, Z., Liu, L., Jia, Z., Liu, Y., Gruteser, M., Raychaudhuri, D., & Zhang, Y. (2021) · Proceedings of the 27th Annual International Conference on Mobile Computing and Networking · doi:10.1145/3447993.3448628 · Cited by 172
“The trend shows that our attention LSTM outperforms vanilla LSTM that demonstrates remarkable accuracy in those prediction work [72, 73], by reducing the loss from 0.”
BackgroundOn attention models for human activity recognition
Murahari, V. S., & Plötz, T. (2018) · Proceedings of the 2018 ACM International Symposium on Wearable Computers · doi:10.1145/3267242.3267287 · Cited by 157
“We employ standard training and evaluation protocols based on hold out datasets as they have been defined in the original publications (and summarized, e.g., in [4]).”
BackgroundMethodInfluentialSelf-citationKwon, H., Tong, C., Haresamudram, H., Gao, Y., Abowd, G. D., Lane, N. D., & Plötz, T. (2020) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3411841 · Cited by 152
“available datasets for human activity recognition contain imbalanced classes. For example, approximately 75% of the Opportunity dataset (which has 18 classes in total) [14] consists of the null class [27], making it challenging to design classifiers. The activities being studied also impact the class imbalance to some…”
BackgroundInfluentialSelf-citationContrastive Predictive Coding for Human Activity Recognition
Haresamudram, H., Essa, I., & Plötz, T. (2021) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3463506 · Cited by 150
“Leveraging the temporal correlations in wearable sensor data at modeling level, (variants of) recurrent neural networks (RNNs) have been applied previously [27, 29, 80].”
BackgroundMethodInfluentialSelf-citationA review of machine learning-based human activity recognition for diverse applications
Kulsoom, F., Narejo, S., Mehmood, Z., Chaudhry, H. N., Butt, A., & Bashir, A. K. (2022) · Neural Computing and Applications · doi:10.1007/s00521-022-07665-9 · Cited by 168
Soleimani, E., & Nazerfard, E. (2021) · Neurocomputing · doi:10.1016/j.neucom.2020.10.056 · Cited by 146
SelfHAR: Improving Human Activity Recognition through Self-training with Unlabeled Data
Tang, C. I., Perez-Pozuelo, I., Spathis, D., Brage, S., Wareham, N., & Mascolo, C. (2021) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3448112 · Cited by 140
“In recent years, the use of deep learning for human activity recognition has become more prevalent [11, 17, 34, 35].”
BackgroundA Human-AI Collaborative Approach for Clinical Decision Making on Rehabilitation Assessment
Lee, M. H., Siewiorek, D. P., Smailagic, A., Bernardino, A., & Bermúdez i Badia, S. B. (2021) · Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3411764.3445472 · Cited by 137
“In contrast to most related work of rehabilitation monitoring systems that focuses on improving the performance of monitoring and understanding an activity with a complex deep learning model [19, 34, 46, 49], our work highlights the importance of creating a human-centered, interactive approach”
BackgroundPeng (2018) · doi:10.1145/3214277 · Cited by 134
“Guan and Plötz [8] built various LSTM networks and fused them in score-level to improve HAR performance.”
BackgroundWearable Sensor-Based Human Activity Recognition Using Hybrid Deep Learning Techniques
Wang, H., Zhao, J., Li, J., Tian, L., Tu, P., Cao, T., An, Y., Wang, K., & Li, S. (2020) · Security and Communication Networks · doi:10.1155/2020/2132138 · Cited by 124
“Due to the superior performance of LSTM in behavior recognition application, Guan and Plötz [19] and Qi et al. [20] improved the LSTM and proposed an integration model, integrating different LSTM learners into an integrated classifier.”
BackgroundQian, B., Su, J., Wen, Z., Jha, D. N., Li, Y., Guan, Y., Puthal, D., James, P., Yang, R., Zomaya, A. Y., et al. (2021) · ACM Computing Surveys · doi:10.1145/3398020 · Cited by 132
“An improved RNN named Long Short Term Memory (LSTM) [122], including complex gates and memory cells within the hidden units for “better memories,” became popular in various applications such as speech recognition [104], video analysis [273], language translation [177], activity recognition [108], and so on.”
BackgroundSelf-citationShallow Convolutional Neural Networks for Human Activity Recognition Using Wearable Sensors
Huang, W., Zhang, L., Gao, W., Min, F., & He, J. (2021) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2021.3091990 · Cited by 129
“For example, our method outperforms [16], [32] and [36] by an accuracy improvement of 1.97%, 0.53% and 6.35% respectively.”
MethodInfluentialMultimodal Federated Learning on IoT Data
Zhao, Y., Barnaghi, P. M., & Haddadi, H. (2022) · IEEE/ACM International Conference on Internet-of-Things Design and Implementation (IoTDI) · doi:10.1109/iotdi54339.2022.00011 · Cited by 170
“For training autoencoders on time-series data, we use long short-term memory (LSTM) [49] autoencoders [46] in our experiments for local training and use the bagging strategy [50] to train our models with random batch sizes and sequence lengths.”
MethodHuang, W., Zhang, L., Wu, H., Min, F., & Song, A. (2022) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2022.3174816 · Cited by 155
“Following recent research literatures [15], [21], [39], [40], [41], we build a 3-layer CNN as our baseline, which could provide very close results comparing to representative convolutional networks.”
MethodOneFi: One-Shot Recognition for Unseen Gesture via COTS WiFi
Xiao, R., Liu, J., Han, J., & Ren, K. (2021) · ACM Conference on Embedded Networked Sensor Systems (SenSys) · doi:10.1145/3485730.3485936 · Cited by 141
“Traditional HGR solutions are mainly camera-based [12, 24, 43] or device-based [2, 14, 35].”
BackgroundHuman Activity Recognition Based on Motion Sensor Using U-Net
Zhang, Y., Zhang, Z., Zhang, Y., Bao, J., Zhang, Y., & Deng, H. (2019) · IEEE Access · doi:10.1109/access.2019.2920969 · Cited by 121
“Guan Y and Ploetz proposed a HAR model based on Ensembles of deep Long Short Term Memory, which has achieved a good recognition effect on Opportunity, PAMAP2 and Skoda datasets [29].”
A Multichannel CNN-GRU Model for Human Activity Recognition
Lu, L., Zhang, C., Cao, K., Deng, T., & Yang, Q. (2022) · IEEE Access · doi:10.1109/access.2022.3185112 · Cited by 124
“…in feedforward neural network (FNN), and is widely used in the fields of sequence annotation, image annotation, etc. Long short-term memory (LSTM) [21] and GRU [22], two variants of RNNs, are used to solve the gradient disappearance and gradient explosion problems of RNNs. Compared with LSTM, GRU…”
MethodA Comprehensive Survey on Deep Learning Methods in Human Activity Recognition
Kaseris, M., Kostavelis, I., & Malassiotis, S. (2024) · Machine Learning and Knowledge Extraction · doi:10.3390/make6020040 · Cited by 128
Yao, R., Lin, G., Shi, Q., & Ranasinghe, D. C. (2018) · Pattern Recognition · doi:10.1016/j.patcog.2017.12.024 · Cited by 105
FedHAR: Semi-Supervised Online Learning for Personalized Federated Human Activity Recognition
Yu, H., Chen, Z., Zhang, X., Chen, X., Zhuang, F., Xiong, H., & Cheng, X. (2023) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2021.3136853 · Cited by 113
“For instance, some deep learning techniques, such as recurrent neural network (RNN) based methods [1], [5] or convolution neural network (CNN) based methods [2], [6], which are usually trained offline in a centralized way.”
BackgroundMethodColloSSL: 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
“Later, deep convolutional and recurrent networks [18, 19, 64] were shown to be effective in learning feature extractors from labeled data.”
BackgroundExtracting biological age from biomedical data via deep learning: too much of a good thing?
Pyrkov, T. V., Slipensky, K., Barg, M., Kondrashin, A., Zhurov, B., Zenin, A., Pyatnitskiy, M., Menshikov, L., Markov, S., & Fedichev, P. O. (2018) · Scientific Reports · doi:10.1038/s41598-018-23534-9 · Cited by 111
“However, deep CNNs bring the idea to an entirely new level and are particularly useful for automation of the most relevant features engineering from the data in relation to human activity recognition (HAR) [33, 34], spe-”
BackgroundHuman Action Recognition Using Deep Learning Methods on Limited Sensory Data
Tüfek, N., Yalcin, M., Altıntaş, M., Kalaoğlu, F., Li, Y., & Bahadır, S. K. (2019) · IEEE Sensors Journal · doi:10.1109/jsen.2019.2956901 · Cited by 117
“There are plenty studies that take advantage of interpreting capability of sequential data for activity recognition by using wearables [38], [39].”
BackgroundZebin, T., Sperrin, M., Peek, N., & Casson, A. J. (2018) · 2018 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) · doi:10.1109/embc.2018.8513115 · Cited by 107
Abedin, A., Ehsanpour, M., Shi, Q., Rezatofighi, H., & Ranasinghe, D. C. (2021) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3448083 · Cited by 107
“3 10-4 3 10-3 3 10-1 3 10-1 Table 2: A comparison of sample-wise activity recognition performance based on class-averaged f1-scores on the holdout test sequences. The baseline results are quoted from [37, 15], except for (*) where the published code is used with our evaluation protocol. HAR Model Opportunity PAMAP2 Sk…”
BackgroundMethodInfluentialSuh, S., Rey, V. F., & Lukowicz, P. (2023) · Knowledge-Based Systems · doi:10.1016/j.knosys.2022.110143 · Cited by 107
Human skeleton pose and spatio-temporal feature-based activity recognition using ST-GCN
Lovanshi, M., & Tiwari, V. (2023) · Multimedia Tools and Applications · doi:10.1007/s11042-023-16001-9 · Cited by 112
Mohammadian Rad, N., Kia, S. M., Zarbo, C., van Laarhoven, T., Jurman, G., Venuti, P., Marchiori, E., & Furlanello, C. (2018) · Signal Processing · doi:10.1016/j.sigpro.2017.10.011 · Cited by 94
Dilated causal convolution with multi-head self attention for sensor human activity recognition
Hamad, R. A., Kimura, M., Yang, L., Woo, W. L., & Wei, B. W. (2021) · Neural Computing and Applications · doi:10.1007/s00521-021-06007-5 · Cited by 109
Deep Learning for Human Activity Recognition in Mobile Computing
Plotz, T., & Guan, Y. (2018) · Computer · doi:10.1109/mc.2018.2381112 · Cited by 99
Self-citationJafari, A., Ganesan, A., Thalisetty, C. S. K., Sivasubramanian, V., Oates, T., & Mohsenin, T. (2018) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2018.2848647 · Cited by 103
Cross-Dataset Activity Recognition via Adaptive Spatial-Temporal Transfer Learning
Qin, X., Chen, Y., Wang, J., & Yu, C. (2019) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3369818 · Cited by 87
“Some existing methods achieve HAR by exploiting the deep understanding of signal data [61], or focusing on machine learning methods such as traditional methods of active learning [25], similarity graph [43] and deep learning methods [14, 22, 24, 31, 53].”
MethodZhao, J., & Obonyo, E. (2020) · Advanced Engineering Informatics · doi:10.1016/j.aei.2020.101177 · Cited by 98
WEAR: An Outdoor Sports Dataset for Wearable and Egocentric Activity Recognition
Bock, M., Kuehne, H., Laerhoven, K. V., & Moeller, M. (2024) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3699776 · Cited by 124
Exploring cybersecurity issues in 5G enabled electric vehicle charging station with deep learning
Basnet, M., & Ali, M. H. (2021) · IET Generation Transmission & Distribution · doi:10.1049/gtd2.12275 · Cited by 92
FedDL: Federated Learning via Dynamic Layer Sharing for Human Activity Recognition
Tu, L., Ouyang, X., Zhou, J., He, Y., & Xing, G. (2021) · Proceedings of the 19th ACM Conference on Embedded Networked Sensor Systems · doi:10.1145/3485730.3485946 · Cited by 87
On the role of features in human activity recognition
Haresamudram, H., Anderson, D. V., & Plötz, T. (2019) · Proceedings of the 23rd International Symposium on Wearable Computers · doi:10.1145/3341163.3347727 · Cited by 90
“The wearables community also has adopted such techniques and remarkable recognition results on benchmark dataset can now be achieved–offline–using variants of end-to-end learning methods, e.g., [15, 34, 46].”
BackgroundMethodSelf-citationSimilarity-based deep learning approach for remaining useful life prediction
Hou, M., Pi, D., & Li, B. (2020) · Measurement · doi:10.1016/j.measurement.2020.107788 · Cited by 88
Interpretable and Accurate Convolutional Neural Networks for Human Activity Recognition
Kim, E. (2020) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2020.2972628 · Cited by 84
Cheng, X., Zhang, L., Tang, Y., Liu, Y., Wu, H., & He, J. (2022) · IEEE Sensors Journal · doi:10.1109/jsen.2022.3149337 · Cited by 82
Semi-supervised convolutional neural networks for human activity recognition
Zeng, M., Yu, T., Wang, X., Nguyen, L. T., Mengshoel, O. J., & Lane, I. (2017) · 2017 IEEE International Conference on Big Data (Big Data) · doi:10.1109/bigdata.2017.8257967 · Cited by 85
“In order to capture the temporal dependencies of the sensor data, deep recurrent networks, especially long short-term memory cells (LSTMs), have achieved promising performance in HAR [3], [14].”
BackgroundA Survey on Anomalous Behavior Detection for Elderly Care Using Dense-Sensing Networks
Deep, S., Zheng, X., Karmakar, C. K., Yu, D., Hamey, L., & Jin, J. (2019) · IEEE Communications Surveys & Tutorials · doi:10.1109/comst.2019.2948204 · Cited by 88
“In addition to the solutions based on traditional machine learning methods, recent studies start using deep learning methods for HAR such as Convolutional Neural Network (CNN) [123], [124] or Recurrent Neural Network (RNN) [125], [126], which represented a better performance than traditional machine learning methods i…”
MethodZhai, B., Perez-Pozuelo, I., Clifton, E. A. D., Palotti, J., & Guan, Y. (2020) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3397325 · Cited by 96
“In [18], Guan and Ploetz developed a LSTM ensemble model via epoch-wise bagging for efficient training.”
MethodSelf-citationNeural Network Ensembles for Sensor-Based Human Activity Recognition Within Smart Environments
Irvine, N., Nugent, C., Zhang, S., Wang, H., & Ng, W. W. Y. (2019) · Sensors · doi:10.3390/s20010216 · Cited by 87
“For example, [52–54] proposed ensemble deep learning techniques for HAR, which revealed positive results and robustness.”
BackgroundAttention-Based Deep Learning Framework for Human Activity Recognition With User Adaptation
Buffelli, D., & Vandin, F. (2021) · IEEE Sensors Journal · doi:10.1109/jsen.2021.3067690 · Cited by 68
Human Activity Recognition with Smartphone Inertial Sensors Using Bidir-LSTM Networks
Yu, S., & Qin, L. (2018) · International Conference on Mechanical, Control and Computer Engineering (ICMCCE) · doi:10.1109/icmcce.2018.00052 · Cited by 80
Improving Deep Learning for HAR with Shallow LSTMs
Bock, M., Hölzemann, A., Moeller, M., & Van Laerhoven, K. (2021) · 2021 International Symposium on Wearable Computers · doi:10.1145/3460421.3480419 · Cited by 74
“suggested different architectures [4, 10–12, 22, 26, 30, 40–42], evaluated the generality of architectures [13] and assessed the applicability in real-world scenarios [9, 12, 16, 21].”
BackgroundLi, B., Yao, Z., Wang, J., Wang, S., Yang, X., & Sun, Y. (2020) · Electronics · doi:10.3390/electronics9111919 · Cited by 74
“demonstrated that ensembles of deep LSTM learners outperform individual LSTM networks in human activity recognition using wearables [36].”
BackgroundMethodInfluentialTrends in human activity recognition with focus on machine learning and power requirements
Nguyen, B., Coelho, Y., Bastos-Filho, T., & Krishnan, S. (2021) · Machine Learning with Applications · doi:10.1016/j.mlwa.2021.100072 · Cited by 93
Alessandrini, M., Biagetti, G., Crippa, P., Falaschetti, L., & Turchetti, C. (2021) · Electronics · doi:10.3390/electronics10141715 · Cited by 79
A study of deep neural networks for human activity recognition
Sansano, E., Montoliu, R., & Belmonte Fernández, Ó. (2020) · Computational Intelligence · doi:10.1111/coin.12318 · Cited by 73
“LSTMs are the most widely RNN architectures used today to process sequential inputs like speech and language.”
InfluentialHuman activity recognition using a multi-branched CNN-BiLSTM-BiGRU model
Lalwani, P. D., & Ganeshan, R. (2024) · Applied Soft Computing · doi:10.1016/j.asoc.2024.111344 · Cited by 76
Outlier Detection in Wearable Sensor Data for Human Activity Recognition (HAR) Based on DRNNs
Munoz-Organero, M. (2019) · IEEE Access · doi:10.1109/access.2019.2921096 · Cited by 69
Barut, O., Zhou, L., & Luo, Y. (2020) · IEEE Internet of Things Journal · doi:10.1109/jiot.2020.2996578 · Cited by 70
“Later, the state-of-the-art results with three benchmark data sets (Opportunity, PAMAP2, and Skoda) are obtained by Guan and Plotz [18] who tackled HAR problems with ensembles of the deep LSTM network.”
BackgroundMulti-sensor fusion based on multiple classifier systems for human activity identification
Nweke, H. F., Wah, T. Y., Mujtaba, G., Alo, U. R., & Al-Garadi, M. A. (2019) · Human-centric Computing and Information Sciences · doi:10.1186/s13673-019-0194-5 · Cited by 76
“Guan and Plötz [55] implemented epoch bagging method for human activity identification by utilizing probabilities selection of the subset of the original data for mini-batch based training of Long Short Term Memory and stochastic based gradient descent learning.”
MethodDeep learning based multimodal complex human activity recognition using wearable devices
Chen, L., Liu, X., Peng, L., & Wu, M. (2020) · Applied Intelligence · doi:10.1007/s10489-020-02005-7 · Cited by 72
“Guan et al. [25] integrated various deep learning models to improve human activity recognition accuracy.”
BackgroundLv, M., Xu, W., & Chen, T. (2019) · Neurocomputing · doi:10.1016/j.neucom.2019.06.051 · Cited by 64
(2018) · International Journal of Machine Learning and Computing · doi:10.18178/ijmlc.2018.8.6.748 · Cited by 57
Dual-Branch Interactive Networks on Multichannel Time Series for Human Activity Recognition
Tang, Y., Zhang, L., Wu, H., He, J., & Song, A. (2022) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2022.3193148 · Cited by 64
“Instead of using convolutional architecture, [12] ensembles multiple Long Short-Term Memory (LSTM) classifiers for sensor-based HAR, showing superior performance on several public HAR benchmark datasets.”
BackgroundMethodSemi-supervised Federated Learning for Activity RecognitionPreprint
Zhao, Y., Liu, H., Li, H., Barnaghi, P. M., & Haddadi, H. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.00851 · Cited by 76
“Many deep learning models such as long short-term memory (LSTM) [13– 15] or convolutional neural network (CNN) [14] can be applied at the edge for HAR.”
BackgroundMethodLei, Y., Karimi, H. R., & Chen, X. (2022) · Neurocomputing · doi:10.1016/j.neucom.2022.06.080 · Cited by 62
Mahmud, T., Sazzad Sayyed, A. Q. M., Fattah, S. A., & Kung, S.-Y. (2021) · IEEE Sensors Journal · doi:10.1109/jsen.2020.3015781 · Cited by 58
Application of ensemble RNN deep neural network to the fall detection through IoT environment
Farsi, M. A. (2020) · Alexandria Engineering Journal · doi:10.1016/j.aej.2020.06.056 · Cited by 60
On the Challenges and Potential of Using Barometric Sensors to Track Human Activity
Manivannan, A., Chin, W. C. B., Barrat, A., & Bouffanais, R. (2020) · Sensors · doi:10.3390/s20236786 · Cited by 63
“The problem of class imbalance in datasets—the fact that some activities are more prevalent than others and are, therefore, over-represented—is a recurrent condition found in both public and private datasets and known to affect recognition accuracy [47,52,53].”
BackgroundHuman activity recognition: A comprehensive review
Kaur, H., Rani, V., & Kumar, M. (2024) · Expert Systems · doi:10.1111/exsy.13680 · Cited by 67
IF-ConvTransformer: A Framework for Human Activity Recognition Using IMU Fusion and ConvTransformer
Zhang, Y., Wang, L., Chen, H., Tian, A., Zhou, S., & Guo, Y. (2022) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3534584 · Cited by 46
“To this end, some classical RNNs [8, 14], e.g., Long Short-Term Memory (LSTM) and Gate Recurrent Unit (GRU), were introduced to HAR.”
MethodRDeepSense: Reliable Deep Mobile Computing Models with Uncertainty Estimations
Yao, S., Zhao, Y., Shao, H., Zhang, A., Zhang, C., Li, S., & Abdelzaher, T. (2018) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3161181 · Cited by 52
“These specifically designed models have shown significant improvements on audio sensing [35], tracking and localization [55], human activity recognition [13, 25, 44, 55], and user identification tasks [55].”
BackgroundA study on IMU-Based Human Activity Recognition Using Deep Learning and Traditional Machine Learning
Hou, C. (2020) · International Conference on Computer and Communication Systems (ICCCS) · doi:10.1109/icccs49078.2020.9118506 · Cited by 56
“The Skoda dataset is a dataset covering the problem of recognizing activities of assembly-line workers inside a car production environment.[10].”
BackgroundDomain Generalization for Activity Recognition via Adaptive Feature Fusion
Qin, X., Wang, J., Chen, Y., Lu, W., & Jiang, X. (2023) · ACM Transactions on Intelligent Systems and Technology · doi:10.1145/3552434 · Cited by 55
“Many machine learning methods have been adopted to improve the performance of HAR, such as Support Vector Machines (SVM), K-Nearest Neighbor (KNN), Random Forest (RF), and the deep learning models including Convolutional Neural Networks (CNN) [25], Long-short Term Memory (LSTM) [21] and others [53].”
MethodCheng, D., Zhang, L., Bu, C., Wu, H., & Song, A. (2023) · Knowledge-Based Systems · doi:10.1016/j.knosys.2023.110789 · Cited by 55
Joint Learning of Temporal Models to Handle Imbalanced Data for Human Activity Recognition
Hamad, R. A., Yang, L., Woo, W. L., & Wei, B. W. (2020) · Applied Sciences · doi:10.3390/app10155293 · Cited by 61
“Therefore, robust learners LSTM and 1D CNN for temporal data are used, which have high variance and low bias due to their almost universal function approximation ability [45] for delivering the joint learning recognition.”
MethodWeakly Supervised Human Activity Recognition From Wearable Sensors by Recurrent Attention Learning
He, J., Zhang, Q., Wang, L., & Pei, L. (2019) · IEEE Sensors Journal · doi:10.1109/jsen.2018.2885796 · Cited by 56
“Ha and Choi [33] also demonstrated that ConvNets (2D convolutions) present better results compared with 1D convolutions (Conv1D) [30], RNN [34] or Long Short Term Memory networks (LSTM) [35].”
ResultDeActive: Scaling Activity Recognition with Active Deep Learning
Hossain, H. M. S., Al Haiz Khan, M. A., & Roy, N. (2018) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3214269 · Cited by 56
“Guan et al. [33] proposed a Long Short Term LSTM based ensemble model and tackled the problems of having imbalanaced and problematic data for wearable devices.”
Background
Metadata
- Authors
- Guan, Y., & Plötz, T.
- Affiliations
- Newcastle University (GB)
- Georgia Institute of Technology (US)
- ORCID
- Guan, Yu
- Venue
- Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Published
- 2017-06-30
- Volume
- 1
- Issue
- 2
- Pages
- 1-28
- ISSN
- 2474-9567 (ISSN-L)
- OpenAlex
- W2604630936
- Cited by
- 511
- References
- 65
- Citations by year
- 2026: 33 · 2025: 35 · 2024: 59 · 2023: 60 · 2022: 70 · 2021: 82 · 2020: 87 · 2019: 46 · 2018: 33 · 2017: 6
- FWCI
- 19.53
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- Engineering and Physical Sciences Research Council (EP/M020576/1)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Context-Aware Activity Recognition Systems
- Physical Sciences › Computer Science › Human-Computer Interaction › Hand Gesture Recognition Systems
- Physical Sciences › Engineering › Biomedical Engineering › Gait Recognition and Analysis
- Keywords
- human activity recognition, wearable sensors, deep learning, ensemble classifiers, class imbalance, data quality, training protocols, PAMAP2, Opportunity dataset
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.
A language model drew the claims from the paper on 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
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.