The DOI matches its Crossref record; 95 of the 95 references it cites were found in a registry or an index; 27 of the 50 sources its claims cite were checked for support.
Assessing the State of Self-Supervised Human Activity Recognition Using Wearables
Open accessAbstract
The emergence of self-supervised learning in the field of wearables-based human activity recognition (HAR) has opened up opportunities to tackle the most pressing challenges in the field, namely to exploit unlabeled data to derive reliable recognition systems for scenarios where only small amounts of labeled training samples can be collected. As such, self-supervision, i.e., the paradigm of 'pretrain-then-finetune' has the potential to become a strong alternative to the predominant end-to-end training approaches, let alone hand-crafted features for the classic activity recognition chain. 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. With the initial success of these methods, the time has come for a systematic inventory and analysis of the potential self-supervised learning has for the field. This paper provides exactly that. We assess the progress of self-supervised HAR research by introducing a framework that performs a multi-faceted exploration of model performance. We organize the framework into three dimensions, each containing three constituent criteria, such that each dimension captures specific aspects of performance, including the robustness to differing source and target conditions, the influence of dataset characteristics, and the feature space characteristics. We utilize this framework to assess seven state-of-the-art self-supervised methods for HAR, leading to the formulation of insights into the properties of these techniques and to establish their value towards learning representations for diverse scenarios.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
71 Own result · 61 No basis given · 38 Own interpretation · 17 Own method · 15 No full text of the sources · 9 Partly supported · 6 Supported · 1 Not supported · 1 Sources not found · 1 Not checked yet · accuracy not yet measured
The cited sources support only part of this claim — read each source’s verdict below.
In the 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.”
- Also stated in: § 1 INTRODUCTION · plain
“Methods such as Multi-task self-supervision [62], Autoencoders [25], Masked reconstruction [26], and Contrastive Predictive Coding (CPC) [27] have been proposed, which aim towards learning generic representations for human activity recognition.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Abstract
What the sources say
- Supportedcites [62] 62. 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. source
The paper identifies its contribution as multi-task self-supervision for human activity recognition (p. 1; §6, p. 23). It therefore supports the citing sentence’s use of this source as an example of a self-supervised HAR method; the other named methods are not attributed to this source. (LLM, conf=0.93)
- Not checked — no full text of the sourcecites [26] 26. 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. source
- Partly supportedcites [27] 27. Harish Haresamudram, Irfan Essa, and Thomas Plötz. 2021. Contrastive predictive coding for human activity recognition. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 5, 2 (2021), 1-26. sourceExperimental — accuracy not yet measured
Section 2.4 (pp. 4–5) describes multi-task self-supervision and masked reconstruction for HAR, and Section 3 (p. 5) introduces CPC for HAR. Section 5.1 (p. 9) also lists autoencoders as an unsupervised comparison, but the copy does not identify them as self-supervised; a search for SimCLR found no passage identifying it as a HAR method. (LLM, conf=0.88)
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.
- Also stated in: § 1 INTRODUCTION · plain
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“This practice of training first on large amounts of unlabeled data, followed by the fine-tuning on smaller quantities of labeled data has shown great promise for activity recognition applications [26, 27, 62-64, 71, 72].”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 1 INTRODUCTION
What the sources say
- Not checked — no full text of the sourcecites [26] 26. 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. source
- Supportedcites [27] 27. Harish Haresamudram, Irfan Essa, and Thomas Plötz. 2021. Contrastive predictive coding for human activity recognition. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 5, 2 (2021), 1-26. source
Section 6.2.3 (p. 18) reports that pre-training on unlabeled data outperformed fully supervised training when labeled data was limited. It also describes fine-tuning representations learned from large movement datasets for target activity-recognition tasks using labeled data. (LLM, conf=0.94)
- Supportedcites [62] 62. 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. source
The Results (p. 15) report that self-supervised pre-training on unlabeled data, followed by fine-tuning with few labeled instances, significantly improved activity-recognition performance on several datasets. The paper specifies that fine-tuning was limited to the last convolutional layer. (LLM, conf=0.94)
- Not checked — no full text of the sourcecites [63] 63. Aaqib Saeed, Flora D Salim, Tanir Ozcelebi, and Johan Lukkien. 2020. Federated Self-Supervised Learning of Multisensor Representations for Embedded Intelligence. IEEE Internet of Things Journal 8, 2 (2020), 1030-1040. source
- Supportedcites [64] 64. Aaqib Saeed, Victor Ungureanu, and Beat Gfeller. 2021. Sense and Learn: Self-supervision for omnipresent sensors. Machine Learning with Applications (2021), 100152. source
Section 4.3.2 describes pre-training on unlabeled instances, then fine-tuning with 5 or 10 labeled examples per class. The same section reports improved F-scores on multiple activity-recognition datasets compared with training from scratch, supporting the claim that this approach has shown promise. (LLM, conf=0.94)
- Supportedcites [71] 71. Chi Ian Tang, Ignacio Perez-Pozuelo, Dimitris Spathis, Soren Brage, Nick Wareham, and Cecilia Mascolo. 2021. SelfHAR: Improving Human Activity Recognition through Self-training with Unlabeled Data. arXiv preprint arXiv:2102.06073 (2021). source
The SelfHAR pipeline description (Section 3.2 and Figure 2; page unavailable) says the student model is pre-trained using a large unlabeled dataset and then fine-tuned with labeled training data. Section 5.3 reports consistent performance improvements in human activity recognition when labeled samples are limited. (LLM, conf=0.91)
- Partly supportedcites [72] 72. Chi Ian Tang, Ignacio Perez-Pozuelo, Dimitris Spathis, and Cecilia Mascolo. 2020. Exploring Contrastive Learning in Human Activity Recognition for Healthcare. arXiv preprint arXiv:2011.11542 (2020). sourceExperimental — accuracy not yet measured
The Introduction (p. 2) says contrastive representations improved activity-recognition performance when fine-tuned with labels, and the Evaluation Results (p. 4) report a gain over fully supervised models. The paper supports the promise of unlabeled pre-training followed by labeled fine-tuning, but it does not establish the claimed comparison between large amounts of unlabeled data and smaller quantities of labeled data. (LLM, conf=0.83)
plainconditional- 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.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Effective performance across differing sensor locations allows for leveraging large scale unlabeled datasetswhich are typically collected at the wrist (such as Capture-24 [4,18,81], Fenland [46,51], and Biobank [14]), to jumpstart recognition for rarer leg-based activities such as classifying freeze of gait.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 3.1 Robustness to Differing Source and Target Conditions
What the sources say
- Not checked — no full text of the sourcecites [4] 4. S Chan Chang and A Doherty. 2021. Capture-24: Activity tracker dataset for human activity recognition. (2021). source
- Supportedcites [18] 18. Jonathan Gershuny, Teresa Harms, Aiden Doherty, Emma Thomas, Karen Milton, Paul Kelly, and Charlie Foster. 2020. Testing self-report time-use diaries against objective instruments in real time. Sociological Methodology 50, 1 (2020), 318-349. source
The Objectives (p. 1) identify this paper as reporting the CAPTURE-24 project, and the accelerometer description (p. 8) says participants wore the device on their wrist. The citation is attached to CAPTURE-24 as an example of wrist-collected data; the sentence does not attribute the proposed cross-location recognition method to this paper. (LLM, conf=0.88)
- Supportedcites [81] 81. Matthew Willetts, Sven Hollowell, Louis Aslett, Chris Holmes, and Aiden Doherty. 2018. Statistical machine learning of sleep and physical activity phenotypes from sensor data in 96,220 UK Biobank participants. Scientific reports 8, 1 (2018), 1-10. source
The Methods (Participants and Accelerometer; page unavailable) identify UK Biobank as a large dataset of wrist-worn accelerometer recordings, which is the part of the sentence cited to this source. The cross-location transfer and freeze-of-gait proposal is the citing authors’ own argument, not a finding attributed to this paper. (LLM, conf=0.92)
- Could not verifycites [46] 46. Luca A Lotta, Laura BL Wittemans, Verena Zuber, Isobel D Stewart, Stephen J Sharp, Jian'an Luan, Felix R Day, Chen Li, Nicholas Bowker, Lina Cai, et al. 2018. Association of genetic variants related to gluteofemoral vs abdominal fat distribution with type 2 diabetes, coronary disease, and cardiovascular risk factors. Jama 320, 24 (2018), 2553-2563. sourceExperimental — accuracy not yet measured
The Study Design and Outcomes passages describe genetic analyses and DEXA fat-mass measurements using UK Biobank and Fenland data. Our copy does not address wrist-worn sensor datasets, performance across sensor locations, or freeze-of-gait recognition; searches for wrist sensors and freeze of gait found no supporting passage. No page numbers were available in our copy. (LLM, conf=0.75)
- Partly supportedcites [51] 51. Laura O'Connor, Soren Brage, Simon J Griffin, Nicholas J Wareham, and Nita G Forouhi. 2015. The cross-sectional association between snacking behaviour and measures of adiposity: the Fenland Study, UK. British journal of nutrition 114, 8 (2015), 1286-1293. sourceExperimental — accuracy not yet measured
The Methods (Study population) identify Fenland as a large study with data on 10,452 participants, and the Methods (Physical activity and sedentary behaviour covariates) report six days of movement-sensor measurements. Our copy does not establish that these data were collected at the wrist or were unlabeled; it also does not address using them to recognize leg-based activities or freezing of gait. (LLM, conf=0.84)
- Supportedcites [14] 14. Aiden Doherty, Dan Jackson, Nils Hammerla, Thomas Plötz, Patrick Olivier, Malcolm H Granat, Tom White, Vincent T Van Hees, Michael I Trenell, Christoper G Owen, et al. 2017. Large scale population assessment of physical activity using wrist worn accelerometers: the UK biobank study. PloS one 12, 2 (2017), e0169649. source
The Discussion reports wrist-worn accelerometer data from 103,578 UK Biobank participants, and Methods > Accelerometer & Data Collection describes the collection of raw acceleration data. The citation identifies Biobank as an example of a large-scale wrist-sensor dataset; the proposed use for freeze-of-gait recognition is the citing authors’ own application, not a finding attributed to this source. (LLM, conf=0.90)
plaindirect- 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.
Every source the paper cites for this claim was checked and supports it.
In the paper
“Tang et al. [71] find that balanced subsets lead to best downstream recognition performance.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 2.1 Self-Supervised Learning in Human Activity Recognition
What the sources say
- Supportedcites [71] 71. Chi Ian Tang, Ignacio Perez-Pozuelo, Dimitris Spathis, Soren Brage, Nick Wareham, and Cecilia Mascolo. 2021. SelfHAR: Improving Human Activity Recognition through Self-training with Unlabeled Data. arXiv preprint arXiv:2102.06073 (2021). source
The Case Study reports that, among three Fenland pre-training subsets tested on a downstream activity-recognition task, the balanced subset yielded the best performance. It also says a balanced activity distribution gave greater improvement than the low-intensity or active subsets. (LLM, conf=0.93)
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“As demonstrated in [72], this intuition applies to time-series wearable sensor data as well, resulting in strong representations on Motionsense.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 2.1 Self-Supervised Learning in Human Activity Recognition
What the sources say
- Supportedcites [72] 72. Chi Ian Tang, Ignacio Perez-Pozuelo, Dimitris Spathis, and Cecilia Mascolo. 2020. Exploring Contrastive Learning in Human Activity Recognition for Healthcare. arXiv preprint arXiv:2011.11542 (2020). source
The Evaluation Results (p. 4) report that SimCLR was evaluated on MotionSense and achieved a best fine-tuned F1 score of 0.942, up to 0.020 above fully supervised models. The Discussion (p. 4) describes the results as promising. The performance depended on the data transformations used, but the claim does not say that every configuration performed strongly. (LLM, conf=0.91)
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether 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
“As in the original paper for CKA [39], the similarity generally reduces as the depth of the encoder layer increases (see the diagonal of the similarity matrices for reference).”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.3.1 Similarity to Supervised Learning.
What the sources say
- Partly supportedcites [39] 39. Simon Kornblith, Mohammad Norouzi, Honglak Lee, and Geoffrey Hinton. 2019. Similarity of neural network representations revisited. In International Conference on Machine Learning. PMLR, 3519-3529. sourceExperimental — accuracy not yet measured
Section 6.3 (p. 8) reports that networks trained on different datasets have similar representations in their early layers, lending some support to an early-versus-late distinction. But the Transformer-encoder analysis in Appendix F.1 (p. 18) describes similarities between corresponding sublayers without stating that similarity generally decreases with encoder depth; that specific trend is not established by the text of our copy. (LLM, conf=0.76)
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption: The gold-standard for activity recognition generally lies in supervised learning, resulting in the best possible performance.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“More specifically, the earlier layers of the self-supervised network produce very similar representations to supervised learning (a trend also observed in [39]).”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.3.1 Similarity to Supervised Learning.
What the sources say
- Partly supportedcites [39] 39. Simon Kornblith, Mohammad Norouzi, Honglak Lee, and Geoffrey Hinton. 2019. Similarity of neural network representations revisited. In International Conference on Machine Learning. PMLR, 3519-3529. sourceExperimental — accuracy not yet measured
Section 6.3 (p. 8) reports that networks trained on CIFAR-10 and CIFAR-100 develop similar representations in their early layers. That supports the general early-layer similarity trend, but the comparison is across datasets, not between self-supervised and supervised learning. (LLM, conf=0.78)
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption: The gold-standard for activity recognition generally lies in supervised learning, resulting in the best possible performance.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“One drawback includes its relatively poor performance on multi-sensor datasets [26,27] as the data transformations are specifically designed for accelerometry and cannot be trivially applied to other sensors, for example, gyroscopes and magnetometers, which are typically part of the IMU setups.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
What the sources say
- Not checked — no full text of the sourcecites [26] 26. 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. source
- Partly supportedcites [27] 27. Harish Haresamudram, Irfan Essa, and Thomas Plötz. 2021. Contrastive predictive coding for human activity recognition. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 5, 2 (2021), 1-26. sourceExperimental — accuracy not yet measured
Section 2.4 (p. 4) says multi-task self-supervision uses eight data transformations and accelerometer signals, whereas the cited paper’s approach is sensor-modality agnostic. It does not say that the transformations cannot readily be applied to gyroscopes or magnetometers, or that this causes relatively poor performance on multi-sensor datasets. (LLM, conf=0.84)
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“In a similar trend, self-supervision shows similar performance to all supervised baselines for USC-HAD.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.1.1 Performance across Different Sensor Positions.
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.
- Rests on an assumption: The target datasets' activity labels are accurate and reliable.
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
“they are not only effective across differing sensor positions”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 1 INTRODUCTION
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.
- It may state two separate claims as one. 81
- Probably not a substantial claim. 51
- The quote only partly states this claim. 1
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“We utilize the test set mean F1-score (i.e., macro F1-score) as the main metric to evaluate performance.This is motivated by the substantial class imbalance present in the target datasets (see Fig. 15 for reference), and thus we require a metric that is resistant to such bias in the class distribution [56].”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 4.5 Performance Metric
What the sources say
- Disputedcites [56] 56. David MW Powers. 2020. Evaluation: from precision, recall and F-measure to ROC, informedness, markedness and correlation. arXiv preprint arXiv:2010.16061 (2020). sourceExperimental — accuracy not yet measured
The source makes the contrary point about F1: the Introduction (p. 1) says F-measure propagates underlying prevalence and bias, and “ROC and PN Analyses” (p. 5) shows F1 varies with prevalence and bias. Its multiclass simulation caption (p. 9) also describes harmonic-mean F1 as a biased traditional measure. The source does not assess the citing paper’s target datasets. (LLM, conf=0.86)
Checked by a reasoning model — may be wrong.
plaindirect- 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.
It may state two separate claims as one.
- Disputed
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Therefore, each target dataset results in its own pre-trained model, which limits generalizability and thus does not allow for the self-supervised transfer learning setup utilized in our study.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 4.2 Self-supervised HAR Methods
What the sources say
- Partly supportedcites [71] 71. Chi Ian Tang, Ignacio Perez-Pozuelo, Dimitris Spathis, Soren Brage, Nick Wareham, and Cecilia Mascolo. 2021. SelfHAR: Improving Human Activity Recognition through Self-training with Unlabeled Data. arXiv preprint arXiv:2102.06073 (2021). sourceExperimental — accuracy not yet measured
Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
plaindirect- 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.
It may state two separate claims as one.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Capture-24 [4, 18, 81] Wrist 151 -Free living”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Dataset
What the sources say
- Not checked — no full text of the sourcecites [4] 4. S Chan Chang and A Doherty. 2021. Capture-24: Activity tracker dataset for human activity recognition. (2021). source
- Partly supportedcites [18] 18. Jonathan Gershuny, Teresa Harms, Aiden Doherty, Emma Thomas, Karen Milton, Paul Kelly, and Charlie Foster. 2020. Testing self-report time-use diaries against objective instruments in real time. Sociological Methodology 50, 1 (2020), 318-349. sourceExperimental — accuracy not yet measured
The Axivity AX3 band accelerometer section (p. 8) says participants wore the device on their wrist during ordinary activities, supporting the wrist-worn, free-living description. But Sample and setting (p. 6) gives an original sample of 148 participants, not 151, with 124 complete diary, camera and accelerometer records. (LLM, conf=0.90)
- Partly supportedcites [81] 81. Matthew Willetts, Sven Hollowell, Louis Aslett, Chris Holmes, and Aiden Doherty. 2018. Statistical machine learning of sleep and physical activity phenotypes from sensor data in 96,220 UK Biobank participants. Scientific reports 8, 1 (2018), 1-10. sourceExperimental — accuracy not yet measured
The Methods (Participants and Accelerometer; page unavailable) describe CAPTURE-24 participants wearing wrist-worn accelerometers, and the paper describes their activity data as free-living. The participant count does not match: 143 were recruited, and 132 remained for classifier development, not 151. (LLM, conf=0.91)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Self-supervised learning was first explored for human activity recognition in a multi-task setting [62].”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 2.1 Self-Supervised Learning in Human Activity Recognition
What the sources say
- Partly supportedcites [62] 62. 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 paper describes using self-supervised learning for human activity recognition with a multi-task network (p. 1; Conclusions, p. 23). It calls the approach novel, but the cited passages do not establish that it was the first exploration of self-supervised learning for human activity recognition. (LLM, conf=0.85)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“The fourth layer of CPC is the Gated Recurrent Unit [8], which also learns representations similar to supervised learning.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.3.1 Similarity to Supervised Learning.
What the sources say
- Supportedcites [8] 8. Junyoung Chung, Caglar Gulcehre, KyungHyun Cho, and Yoshua Bengio. 2014. Empirical evaluation of gated recurrent neural networks on sequence modeling. arXiv preprint arXiv:1412.3555 (2014). source
The citation identifies the Gated Recurrent Unit. Section 3.2 (p. 4) describes the GRU and its gating mechanism; the statement that it is CPC’s fourth layer concerns the citing authors’ own model, not this source. (LLM, conf=0.90)
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: The gold-standard for activity recognition generally lies in supervised learning, resulting in the best possible performance.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“A strength of this method lies in its applicability and superior performance on multi-sensor data [27], and in scenarios where transition-style activities are present (for example, getting in and out of a car, as in Mobiact).”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
What the sources say
- Supportedcites [27] 27. Harish Haresamudram, Irfan Essa, and Thomas Plötz. 2021. Contrastive predictive coding for human activity recognition. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 5, 2 (2021), 1-26. source
Application Scenarios (pp. 7–8) describes evaluation on datasets using accelerometers and gyroscopes. Activity Recognition (p. 10) reports that CPC outperformed the unsupervised baselines on three datasets and was comparable on the fourth. Analyzing CPC (p. 16) also reports improved recognition of Mobiact transition activities, including getting into and out of a car, with a particular choice of learned features; the performance advantage is not over every comparator or configuration. (LLM, conf=0.87)
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“As mentioned in the original paper [6], the choice of augmentations has a large impact on the pre-training quality.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
What the sources say
- Supportedcites [6] 6. Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A simple framework for contrastive learning of visual representations. In International conference on machine learning. PMLR, 1597-1607. source
Section 3.1 (p. 4) explicitly studies how individual augmentations and their combinations affect learned representations. Section 3.2 (p. 5) reports that stronger color augmentation substantially improves linear evaluation of the learned unsupervised models, supporting the claim that augmentation choices strongly affect pre-training quality. (LLM, conf=0.93)
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“In this work, we follow the philosophy espoused by natural language processing benchmarks such as GLUE [79] and GEM [17], which argues that evaluation on a suite of tasks is necessary for a well-rounded understanding of model performance.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.4 The Necessity for an Evolving Benchmark and a Call to Action
What the sources say
- Supportedcites [79] 79. Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R Bowman. 2018. GLUE: A multi-task benchmark and analysis platform for natural language understanding. arXiv preprint arXiv:1804.07461 (2018). source
The Abstract (p. 1) says useful NLU must work beyond a single task, genre, or dataset and introduces GLUE to evaluate models across diverse tasks. The Tasks section (p. 3) describes nine tasks spanning domains and difficulties, supporting the attributed evaluation philosophy. (LLM, conf=0.92)
- Supportedcites [17] 17. Sebastian Gehrmann, Tosin Adewumi, Karmanya Aggarwal, Pawan Sasanka Ammanamanchi, Aremu Anuoluwapo, Antoine Bosselut, Khyathi Raghavi Chandu, Miruna Clinciu, Dipanjan Das, Kaustubh D Dhole, et al. 2021. The gem benchmark: Natural language generation, its evaluation and metrics. arXiv preprint arXiv:2102.01672 (2021). source
In Dataset Selection (p. 6), GEM explicitly favors diverse high-level tasks over examining a single task in depth. Benchmarks in NLG (p. 3) says unified evaluation across multiple tasks helps compare models, while warning that a single score may mischaracterize performance. Together, these support the attributed philosophy of evaluating across a task suite for a fuller understanding of performance. (LLM, conf=0.89)
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
- Supported
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
“Considering the best performing self-supervised methods such as SimCLR and Multi-task self-supervision, we see significant improvements over classifying on raw data for a majority of the datasets (e.g., on PAMAP2, SimCLR shows an increase of approx.10% F1-score over utilizing raw data).”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.1.1 Performance across Different Sensor Positions.
plaindirect- 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.
It may state two separate claims as one.
- Rests on an assumption: The target datasets' activity labels are accurate and reliable.
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
“Lastly, most of the datasets have class imbalance (see Fig. 15 for reference) with Wetlab, Mobiact, MHEALTH and Myogym having the most severe class compositions.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Dataset
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Pre-training is less effective for Mobiact, with DeepConvLSTM significantly outperforming the best self-supervised methods, increasing by over 10% against CPC and 7% over SimCLR.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.1.1 Performance across Different Sensor Positions.
plaindirect- 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.
It may state two separate claims as one.
- Rests on an assumption: The target datasets' activity labels are accurate and reliable.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The overall trend is that transferring weights from the wrist to the waist results in worsened performance (Mobiact and Motionsense, when compared to the best supervised baseline, i.e., the Conv.classifier) or similar performance (USC-HAD).”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.1.1 Performance across Different Sensor Positions.
plaindirect- 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.
It may state two separate claims as one.
- Rests on an assumption: The target datasets' activity labels are accurate and reliable.
- Rests on an assumption: Sensor data from different datasets are comparable after per-channel normalization, so representations transfer across devices and body locations.
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
“The dataset composition is imbalanced, with around 75% corresponding to either sleep or sit-stand, whereas bicycling comprises of only 0.78% of the train dataset windows, as shown in Fig. 13 in the Appendix.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 4.1.1 Capture-24.
plainn = 135- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“By and large, the LSTM classifier and the MLP classifier show similar performance, with the GRU classifier performing better than both.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.1.1 Performance across Different Sensor Positions.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: The target datasets' activity labels are accurate and reliable.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“As mentioned previously, we observe the highest improvements over DeepConvLSTM, LSTM, GRU, and MLP classifiers with PAMAP2, whereas the Conv.classifier performs comparably.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.1.1 Performance across Different Sensor Positions.
plaincorrelational- 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.
It may state two separate claims as one.
- Rests on an assumption: The target datasets' activity labels are accurate and reliable.
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
“Across both the waist-and leg-based datasets, we note that the performance is substantially lower relative to all supervised learning baselines.Only in the case of SimCLR do we obtain performance comparable to the supervised approaches (Mobiact is an exception).”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.1.1 Performance across Different Sensor Positions.
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.
It may state two separate claims as one.
- Rests on an assumption: The target datasets' activity labels are accurate and reliable.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“In comparison, transferring from the wrist to leg results in comparable performance (Daphnet FoG and PAMAP2, against the Conv.classifier), or gains (MHEALTH).”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.1.1 Performance across Different Sensor Positions.
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.
It may state two separate claims as one.
- Rests on an assumption: The target datasets' activity labels are accurate and reliable.
- Rests on an assumption: Sensor data from different datasets are comparable after per-channel normalization, so representations transfer across devices and body locations.
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
“For Mobiact, the difference between the means of the balanced and imbalanced runs is ∼0.9% when ρ=0.1 (i.e., the rarest class is 10 times smaller than the most frequenct class).This increases to ∼2% and ∼3.3% when ρ is 0.01 and 0.001 respectively.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Source dataset imbalance.
plaindirect- 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.
It may state two separate claims as one.
- Rests on an assumption: Wearables-based movement datasets tend to be imbalanced, with more samples of naturally occurring or easily performed activities than strenuous or rare activities.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“In each case, self-supervision results in significant improvements over DeepConvLSTM for PAMAP2 (∼9% over the mean) and MHEALTH (∼4% over the mean).”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Leg
plaindirect- 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.
It may state two separate claims as one.
- Rests on an assumption: The target datasets' activity labels are accurate and reliable.
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
“In the case of HHAR, which likely has the most overlap with Capture-24, Multi-task self-supervision increases from 51% to 57.5% with the use of the MLP.Similarly, CPC and SimCLR show improvements from around 56% to 58%.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Leg
hedgeddirect- 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.
It may state two separate claims as one.
- Rests on an assumption: The target datasets' activity labels are accurate and reliable.
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
“Multi-task self-supervision and CPC benefit from larger number of participants but there is generally no significant gain between using 25% of the subjects over utilizing the entire training set.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.1 Quantity of the Unlabeled Data Required.
plainn = 135- 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.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The other datasets were recorded at 100 Hz and the reduction in performance occurs to a smaller extent for some of the self-supervised methods.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.1.3 Variations in the Sampling Rate.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: The target datasets' activity labels are accurate and reliable.
- Rests on an assumption: The 50 Hz sampling rate retains sufficient information for the target activities.
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
“For SimSiam and BYOL, providing more pre-training windows slightly improves the F1-score for most target datasets, without showing drastic changes in performance.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Effect of the number of training windows on activity recognition.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
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 note that excellent performance can be obtained while using 10% of the training windows, and in some cases using as few as 1% of the training windows.”
- Also stated in: § Effect of the number of training windows on activity recognition. · plain
“As a general trend, we observe that the self-supervised methods obtain comparable if not better performance when utilizing only 10% (or ∼400k) of training windows.”
- Also stated in: § 6.2 Is the Performance of Self-Supervision Getting Hampered by (Too) Simple Encoder Architectures? · plain
“In our study of the quantity of the unlabeled source dataset windows required for pre-training (see Fig. 4), we observed that using only 10% of the training windows results in comparable if not better downstream activity recognition.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.1 Quantity of the Unlabeled Data Required.
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.
It may state two separate claims as one.
- Also stated in: § Effect of the number of training windows on activity recognition. · plain
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“For other datasets, the best performing self-supervised methods still typically outperform the classifier, although the improvements maybe more modest (e.g., in the case of Daphnet FoG dataset and PAMAP2).”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.3 Performance under Availability of Very Limited Annotated Data.
hedged- 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.
It may state two separate claims as one.
- Rests on an assumption: Very limited labeled data scenarios are of vital importance to activity recognition and can arise due to cost and privacy concerns, such as in medical applications.
- Rests on an assumption: Real-world deployment may allow for acquiring a small labeled dataset from users without significant interruptions, such as during device setup.
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
“Masked Reconstruction leads to a jump in performance between using 1% and 10% of the windows while the F1-score steadies beyond that.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Effect of the number of training windows on activity recognition.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
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
“Surprisingly, using data from just one user does not result in poor performance; rather, the performance is close to utilizing all available training data.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.1 Quantity of the Unlabeled Data Required.
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.
It may state two separate claims as one.
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
“The contrastive learning methods show the highest difference in the train F1-score thereby learning the least generic features, especially SimCLR, SimSiam, and BYOL.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > CPC Autoencoder
plaindirect- 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.
It may state two separate claims as one. The quote only partly states this claim.
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
“The impact of the MLP-based classification is more pronounced on the fine-grained gym activities (e.g., curls, raises, and dips) present in Myogym, with Autoencoder increasing from 11.4% to 35.5%, and BYOL improving from 17% to 39%.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Leg
plaindirect- 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.
It may state two separate claims as one.
- Rests on an assumption: The target datasets' activity labels are accurate and reliable.
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
“In Fig. 10, we see that for the contrastive methods such as CPC, SimCLR, and BYOL, as well as the Autoencoder, around 95% of the variance can be explained from the first five components itself, which increases to over 98% when considering 20 components.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.3.3 Implicit Dimensionality of the Representations.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
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
“the Autoencoder also follows the general trend of higher similarity in the earlier layers with the first two convolutional layers being very similar.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.3.1 Similarity to Supervised Learning.
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: The gold-standard for activity recognition generally lies in supervised learning, resulting in the best possible performance.
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 observe that Multi-task self-supervision generally has a lower difference in the performance between ground truth and randomized labels, indicating that it learns more general representations that enable various downstream scenarios.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > CPC Autoencoder
hedgeddirect- 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.
It may state two separate claims as one.
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
“In general, SimCLR is the most effective self-supervised method for activity recognition using wearables, across both sensor positions and target activities, whereas Multi-task self-supervision also shows high performance.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.3 Performance under Availability of Very Limited Annotated Data.
plaincorrelational- 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.
It may state two separate claims as one.
- Rests on an assumption: Very limited labeled data scenarios are of vital importance to activity recognition and can arise due to cost and privacy concerns, such as in medical applications.
- Rests on an assumption: Real-world deployment may allow for acquiring a small labeled dataset from users without significant interruptions, such as during device setup.
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
“For the lab activities in Wetlab, we observe a drop in performance, with a peak difference of ∼10% between DeepConvLSTM and SimCLR.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Leg
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: The target datasets' activity labels are accurate and reliable.
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
“Myogym follows a similar trend, with reduction ∈ ∼{0.5%, 1.6%, 2.4%} when ρ ∈ {0.1, 0.01, 0.001}.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Source dataset imbalance.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: Wearables-based movement datasets tend to be imbalanced, with more samples of naturally occurring or easily performed activities than strenuous or rare activities.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The main advantages of this method include its simplicity of design, where neither negative pairs nor a momentum encoder or large batch sizes are required for effective learning.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“This method also trains quickly and is not computationally heavy, while also keeping the pre-training hyper-parameter space relatively small.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“These approaches have strong practical value, exhibiting significant improvements in scenarios where labeled data is at a premium.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.5 Lessons Learned and Insights Gained
boosteddirect- 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.
Probably not a substantial claim.
None of the sources cited for this claim have been checked yet.
In the paper
“The main advantages of this method include its simplicity of design, where neither negative pairs nor a momentum encoder or large batch sizes are required for effective learning.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“By and large, a small yet diverse dataset (in terms of participants) can result in better recognition than collecting lots of data from a few participants.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.5 Lessons Learned and Insights Gained
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The target dataset sampling rate must match the source conditions for optimal performance.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.5 Lessons Learned and Insights Gained
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Early layers of the learned encoders show high similarity to supervised learning, whereas the features diverge deeper into the network.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.5 Lessons Learned and Insights Gained
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“A downside to this approach is that it requires considerable parameter tuning, including learning rate and weight decay, future prediction horizon, and batch size.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The model architecture is simple with low training times and a small hyperparameter space.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
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
“For both Multi-task self-supervision and Autoencoder, the performance reduces with the addition of another encoder layer, indicating that such naive increase in encoder depth does not show a corresponding improvement in activity recognition, even if more pre-training data are available.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.2 Is the Performance of Self-Supervision Getting Hampered by (Too) Simple Encoder Architectures?
hedgeddirect- 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.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The training time is considerable, and extends particularly with increasing prediction horizons.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Masked reconstruction has the highest implicit dimensionality and thus more efficiently makes use of the learned representation space.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.5 Lessons Learned and Insights Gained
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“This method also trains quickly and is not computationally heavy, while also keeping the pre-training hyper-parameter space relatively small.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
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
“When the training windows are sub-sampled based on a percentage, the class distribution generally follows the composition of the full train dataset (see Fig. 13) albeit the number of samples in each class are different.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.2 Transformation Functions used for SimCLR, SimSiam, and BYOL
hedgedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Utilizing the means and variances of the source dataset normalization on the target dataset results in considerable performance gains.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.5 Lessons Learned and Insights Gained
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The training itself is relatively fast, while the hyper-parameter space is not very large, comprising of learning rate, weight decay, and the batch size.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
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
“Sleep and sit-stand comprise around 75% of the total dataset”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.2 Transformation Functions used for SimCLR, SimSiam, and BYOL
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
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
“The main advantage of utilizing this approach lies in its effectiveness in learning representations that are applicable for a variety of downstream scenarios, as evidenced by its overall superior performance in Tab. 3 and 4.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
plaindirect- 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.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Self-supervised learning is surprisingly robust to even extreme imbalances in the source dataset, with minimal reduction in performance relative to balanced data.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.5 Lessons Learned and Insights Gained
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: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“While it is computationally light as well, the setup necessitates large batch sizes, as the negative pairs for contrastive learning consist of all other augmented windows in the batch.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
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
“whereas bicycling only consists of 0.78% of the train data.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.2 Transformation Functions used for SimCLR, SimSiam, and BYOL
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Also, layers closer to each other are more similar than those farther apart.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.5 Lessons Learned and Insights Gained
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The training itself is relatively fast, while the hyper-parameter space is not very large, comprising of learning rate, weight decay, and the batch size.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The techniques are data efficient and perform comparably while only utilizing a small fraction of the available data -both from a participant as well as quantity of data standpoints -thereby reducing the computational burden.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.5 Lessons Learned and Insights Gained
plaindirect- 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.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The positives for this technique include its robustness to the choice of augmentations (in direct contrast to SimCLR), simple setup and architecture, and its general effectiveness across a variety of downstream scenarios.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
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 note that excellent performance can be obtained while using 10% of the training windows, and in some cases using as few as 1% of the training windows.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.2 Transformation Functions used for SimCLR, SimSiam, and BYOL
plaindirect- 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.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Given the simplicity of the architecture and the objective, the training time is also significantly shorter relative to other methods, while also being less stressful from a computational perspective.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Multi-task self-supervision succeeds in learning the most generic features of wearable data, and thereby enables wide variety of downstream tasks.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.5 Lessons Learned and Insights Gained
plaindirect- 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: unclear — It is unclear why the paper says it holds.
It may state two separate claims as one.
In the paper
“SimSiam [7] produces surprisingly effective representations with a simple siamese setup.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
What the sources say
- Not checked — no full text of the sourcecites [7] 7. Xinlei Chen and Kaiming He. 2021. Exploring simple siamese representation learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 15750-15758. source
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
In the paper
“Additionally, the Transformer [77] encoder architecture is large and computationally intensive, and has the longest training time by far in this study.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
What the sources say
- Not matchedcites [77] 77. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. Advances in neural information processing systems 30 (2017).
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Not matched
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“The best pre-training combination for each dataset is once again trained with five random seeds, and these trained models are utilized for five randomized classifier runs (using the same seeds as pre-training), in order to account for the effect of the choice of random”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 4.3 Implementation Details
plaindirect- 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: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
In the paper
“whereas Myogym contains 31 fine-grained gym exercises.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Dataset
What the sources say
- Not checked — no full text of the sourcecites [40] 40. Heli Koskimäki, Pekka Siirtola, and Juha Röning. 2017. Myogym: introducing an open gym data set for activity recognition collected using myo armband. In Proceedings of the 2017 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of the 2017 ACM International Symposium on Wearable Computers. 537-546. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
In the paper
“We perform no filtering or denoising on the datasets, as deep networks have shown powerful capabilities towards learning from raw data itself [42].”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 4.4 Data Preparation
What the sources say
- Not checked — no full text of the sourcecites [42] 42. Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. 2015. Deep learning. nature 521, 7553 (2015), 436-444. source
boosteddirect- 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.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Therefore, this evaluation protocol is limited as it only studies one aspect of model performance, which makes it challenging to realize the conditions of effective performance as well as avenues of improvement for these approaches.”
- Also stated in: § 2.1 Self-Supervised Learning in Human Activity Recognition · plain
“Such an evaluation protocol only covers one aspect of measuring the model performance -which is under very similar training and testing conditions.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 1 INTRODUCTION
plaindirect- 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.
It may state two separate claims as one.
- Also stated in: § 2.1 Self-Supervised Learning in Human Activity Recognition · plain
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“We chose three datasets per location-waist/trousers, wrist, and leg/ankle-covering a wide variety of activities and dataset sizes.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 4.1.2 Target Datasets.
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.
Probably not a substantial claim. It may state two separate claims as one.
In the paper
“Overall, every participant appears exactly once as part of the test split and pairwise correlations of analysis windows are effectively eliminated through the user-based splits [22].”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 4.4 Data Preparation
What the sources say
- Not checked — no full text of the sourcecites [22] 22. Nils Hammerla and Thomas Ploetz. 2015. Let's (not) Stick Together: Pairwise Similarity Biases Cross-Validation in Activity Recognition. In Proc. UbiComp. source
plaindirect- 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.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“These self-supervised approaches have the potential to alleviate challenges that come with limited annotations (incl.overfitting, and availability of less-complex modeling approaches only), and are therefore early proponents of the 'pretrain-then-finetune' paradigm shift.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 1 INTRODUCTION
hedgedconditional- 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.
It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“We setup five-fold cross validation for each target dataset.Each fold consists of 20% of randomly chosen users comprising the test split, whereas the remaining 80% of users are once again separated into the train and validation splits at a 80:20 ratio.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 4.4 Data Preparation
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.
Probably not a substantial claim. It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“For each combination, we perform 5-fold cross validation to obtain the average F1-score, which is more robust to the particular choice of participants in the splits”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 4.3 Implementation Details
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
In the paper
“The accuracy and weighted F1-score are not utilized as they are affected by the skewed class distributions, whereas the unweighted mean F1-score, while not ideal, is a reasonable strategy [53].”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 4.5 Performance Metric
What the sources say
- Not checked — no full text of the sourcecites [53] 53. Thomas PlÖtz. 2021. Applying machine learning for sensor data analysis in interactive systems: Common pitfalls of pragmatic use and ways to avoid them. ACM Computing Surveys (CSUR) 54, 6 (2021), 1-25. source
plaindirect- 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.
It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“They have been chosen carefully to cover a wide range of application scenarios, including ranges of sensor positions, activities covered, and sampling rates employed for data recording.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 4 SETUP
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.
Probably not a substantial claim.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“In most of the previously discussed unsupervised and self-supervised approaches including Autoencoders, Multi-task self-supervision, Masked reconstruction, SimCLR, SimSiam, BYOL and CPC, the source and target datasets are typically the same for most experiments, albeit the pre-training does not use label information.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 2.1 Self-Supervised Learning in Human Activity Recognition
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.
It may state two separate claims as one.
A claim is checked only against the cited paper’s full text, never its abstract, and none is available for its sources.
In the paper
“MHEALTH and PAMAP2 both have activities that can be expected in daily living, albeit they are collected at the leg/ankle, rather than the wrist.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Leg
What the sources say
- Not checked — no full text of the sourcecites [2] 2. Oresti Banos, Rafael Garcia, Juan A Holgado-Terriza, Miguel Damas, Hector Pomares, Ignacio Rojas, Alejandro Saez, and Claudia Villalonga. 2014. mHealthDroid: a novel framework for agile development of mobile health applications. In International workshop on ambient assisted living. Springer, 91-98. source
- Not checked — no full text of the sourcecites [61] 61. Attila Reiss and Didier Stricker. 2012. Introducing a new benchmarked dataset for activity monitoring. In 2012 16th international symposium on wearable computers. IEEE, 108-109. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
In the paper
“In some cases, Mobiact [5] is used for pre-training instead as it is has the largest number of participants.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 2.1 Self-Supervised Learning in Human Activity Recognition
What the sources say
- Not checked — no full text of the sourcecites [5] 5. Charikleia Chatzaki, Matthew Pediaditis, George Vavoulas, and Manolis Tsiknakis. 2016. Human daily activity and fall recognition using a smartphone's acceleration sensor. In International Conference on Information and Communication Technologies for Ageing Well and e-Health. Springer, 100-118. source
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“As the Capture-24 dataset is collected at the wrist, we evaluate the learned weights at other on-body locations including the waist/trousers (where people often carry their smartphones), and the leg”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.1.1 Performance across Different Sensor Positions.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
In the paper
“We perform a random search over available combinations for Tab. 3 and 4, as it has been shown to be very effective compared to grid searches [3], while also not consuming extensive computational resources.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 4.3 Implementation Details
What the sources say
- Not checked — no full text of the sourcecites [3] 3. James Bergstra and Yoshua Bengio. 2012. Random search for hyper-parameter optimization. Journal of machine learning research 13, 2 (2012). source
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“It is also interesting to note that the convolutional architecture utilized in many approaches in this study, has long range filters (of sizes 24, 16, and 8), and is likely more apt for the target datasets, as they typically comprise of more static activities such as walking, running, sitting, etc.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.1.1 Performance across Different Sensor Positions.
hedgedconditional- 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: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
In the paper
“For example, in healthcare, collecting large amounts of data may be difficult, due to challenges with participant recruitment.This is compounded by the necessity to enlist healthcare professionals for annotation, which may incur high costs [31].”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 1 INTRODUCTION
What the sources say
- Not checked — no full text of the sourcecites [31] 31. Shruthi K Hiremath and Thomas Plötz. 2020. Deriving effective human activity recognition systems through objective task complexity assessment. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 4, 4 (2020), 1-24. source
hedgedconditional- 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.
It may state two separate claims as one.
In the paper
“For these methods, model pre-training using unlabeled training data is typically either performed utilizing the same dataset as used for training the target activity recognition system-but without using label information-or with a larger background dataset such as Mobiact [5].”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 1 INTRODUCTION
What the sources say
- Not checked — no full text of the sourcecites [5] 5. Charikleia Chatzaki, Matthew Pediaditis, George Vavoulas, and Manolis Tsiknakis. 2016. Human daily activity and fall recognition using a smartphone's acceleration sensor. In International Conference on Information and Communication Technologies for Ageing Well and e-Health. Springer, 100-118. source
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.
It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“In order to facilitate a broad evaluation of these approaches, we curate nine target datasets -with the goal of having a diverse collection of activities, number of participants, sensor locations, dataset sizes, and sensors utilized.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 4.1.2 Target Datasets.
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.
Probably not a substantial claim.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“many wearable sensing datasets are inherently imbalanced, containing larger quantities of more easily performed activities such as walking or lying down, relative to physically demanding movements such as skipping with a rope.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 3.2 Influence of Dataset Characteristics
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: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“in numerous target scenarios, the data collection and subsequent annotation maybe difficult due to cost or privacy issues.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 3.2 Influence of Dataset Characteristics
hedgedconditional- 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: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
In the paper
“Similarly, the Wetlab dataset contains laboratory experiments, which are typically not performed in day-to-day living.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Leg
What the sources say
- Not checked — no full text of the sourcecites [66] 66. Philipp M Scholl, Matthias Wille, and Kristof Van Laerhoven. 2015. Wearables in the wet lab: a laboratory system for capturing and guiding experiments. In Proceedings of the 2015 ACM International Joint Conference on Pervasive and Ubiquitous Computing. 589-599. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As such, this means that the source dataset need not be artificially balanced for effective performance; rather, the unavailability of even coarse labels (for balancing) is not a crutch for learning useful representations.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Source dataset imbalance.
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: unclear — It is unclear why the paper says it holds.
It may state two separate claims as one.
- Rests on an assumption: Wearables-based movement datasets tend to be imbalanced, with more samples of naturally occurring or easily performed activities than strenuous or rare activities.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Using a smaller training is advantageous as it can allow for reduced computation costs and for more comprehensive hyperparameter searching.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.1 Quantity of the Unlabeled Data Required.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Overall, it is preferable to match the source dataset sampling rates as it different sampling rates result in reduced performance.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.1.3 Variations in the Sampling Rate.
plaindirect- 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.
It may state two separate claims as one.
- Rests on an assumption: The 50 Hz sampling rate retains sufficient information for the target activities.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The similarity of the later layers diverges, likely because the supervised representations capture the more activity-specific characterisics of the data, whereas the self-supervised methods optimize different objectives entirely.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.3.1 Similarity to Supervised Learning.
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: The gold-standard for activity recognition generally lies in supervised learning, resulting in the best possible performance.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The pre-training is dependent on reconstructing the source data, and limited exposure to windows of the rarest class renders the model unable to reconstruct them accurately, thereby resulting in less effective representation learning.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Source dataset imbalance.
plaindirect- 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.
It may state two separate claims as one.
- Rests on an assumption: Wearables-based movement datasets tend to be imbalanced, with more samples of naturally occurring or easily performed activities than strenuous or rare activities.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Thus, a majority of the self-supervised methods including Multi-task self-supervision, CPC, Autoencoder, SimCLR, SimSiam, and BYOL do not require the entire training set (135 participants) for best performance, and can be trained for increased efficiency with fewer participants without much loss in performance.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.1 Quantity of the Unlabeled Data Required.
plainn = 135- 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.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Masked reconstruction struggles the most, likely because it depends on surrounding context to predict the missing values.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.1.3 Variations in the Sampling Rate.
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Therefore, self-supervision is more effective when transferring to the wristbased datasets (which match the source dataset location), or the leg.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.3 Performance under Availability of Very Limited Annotated Data.
plaindirect- 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.
It may state two separate claims as one.
- Rests on an assumption: Very limited labeled data scenarios are of vital importance to activity recognition and can arise due to cost and privacy concerns, such as in medical applications.
- Rests on an assumption: Real-world deployment may allow for acquiring a small labeled dataset from users without significant interruptions, such as during device setup.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This is especially true for Masked Reconstruction, which utilizes a large transformer encoder and thus requires more data.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.1 Quantity of the Unlabeled Data Required.
plaindirect- 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.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The comparable performance of all self-supervised methods on such rarely accessible medical data demonstrates their generalization capacity and their value towards utilization in conditions where labeled data maybe difficult to obtain.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Leg
boosteddirect- 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.
It may state two separate claims as one.
- Rests on an assumption: The target datasets' activity labels are accurate and reliable.
- Rests on an assumption: The Capture-24 dataset is representative of free-living human activity, so representations learned on it are useful for other wearable HAR tasks.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“On the other hand, Multi-task self-supervision and Masked reconstruction utilize the latent space more effectively and have higher implicit dimensionality.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.3.3 Implicit Dimensionality of the Representations.
plaindirect- 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.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This is encouraging as the learned representations can generalize to unseen activities and improve performance even when sufficient labeled data is not available.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.3 Performance under Availability of Very Limited Annotated Data.
plaindirect- 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.
It may state two separate claims as one.
- Rests on an assumption: Very limited labeled data scenarios are of vital importance to activity recognition and can arise due to cost and privacy concerns, such as in medical applications.
- Rests on an assumption: Real-world deployment may allow for acquiring a small labeled dataset from users without significant interruptions, such as during device setup.
A conclusion the paper draws from its own results — 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 also note that the Conv.classifier typically performs better than the DeepConvLSTM when training on very limited labels likely due to its relatively small size and simple architecture, leading to less overfitting.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.3 Performance under Availability of Very Limited Annotated Data.
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: Very limited labeled data scenarios are of vital importance to activity recognition and can arise due to cost and privacy concerns, such as in medical applications.
- Rests on an assumption: Real-world deployment may allow for acquiring a small labeled dataset from users without significant interruptions, such as during device setup.
In the paper
“As per the original paper [7], it can be thought of as 'BYOL without the momentum encoder' and 'SimCLR without the negative pairs', thereby acting as the common connection point between several existing methods.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
What the sources say
- Not checked — no full text of the sourcecites [7] 7. Xinlei Chen and Kaiming He. 2021. Exploring simple siamese representation learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 15750-15758. source
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“For each target dataset fold, we randomly select {2, 5, 10, 50, 100} labeled windows per class for training whereas the validation and test splits are untouched.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.3 Performance under Availability of Very Limited Annotated Data.
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.
Probably not a substantial claim. It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“The test-set F1-score obtained after five-fold validation and five randomized runs is visualized in Fig. 11.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.1 Applying Source Dataset Normalization to Target Scenario for Activity Recognition
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Three randomized runs of pre-training (incl.the randomized subsampling) and classification are performed, and the mean and standard deviation of five-fold test set F1-scores are plotted in Fig. 4.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Effect of the number of training windows on activity recognition.
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.
Probably not a substantial claim. It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“We perform three randomized runs for choosing the training windows, performing pre-training and activity recogntition, and plot the mean of the target test set F1-scores along with the standard deviation.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.1 Quantity of the Unlabeled Data Required.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“In practice, we perform five randomized runs for the pre-training and in each case, the most frequent and the rarest classes are chosen randomly, thereby making the aggregated performance resistant the choice of the majority class.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Source dataset imbalance.
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.
Probably not a substantial claim. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As we only utilize data from a single accelerometer for the assessment, the current version of the framework does not take into consideration issues typically arise in multi-sensor datasets, such as variation in data distribution across devices, synchronization etc.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 7 SUMMARY AND CONCLUSION
plaindirect- 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.
It may state two separate claims as one.
In the paper
“The downside to this approach is that it can collapse to trivial solutions, but the stop gradient operation can generally counteract such scenarios.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
What the sources say
- Not checked — no full text of the sourcecites [7] 7. Xinlei Chen and Kaiming He. 2021. Exploring simple siamese representation learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 15750-15758. source
plainconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“We randomly sample 1, 000 training windows in five randomized runs and extract the representations from the last encoder layer.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.3.3 Implicit Dimensionality of the Representations.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Principal Component Analysis (PCA) is performed on the extracted features for each target dataset, and the cumulative fraction of the explained variance is averaged across the five randomized runs.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.3.3 Implicit Dimensionality of the Representations.
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.
Probably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“These results and findings are encouraging as they indicate the broad applicability of these self-supervised methods in wide-ranging target scenarios, and affirm the capability of these techniques for empowering human activity recognition in applications where data both collection and annotation may be challenging to accomplish.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 7 SUMMARY AND CONCLUSION
hedgedconditional- 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.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The disadvantage of utilizing autoencoders is that it may not result the best possible unsupervised performance.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“The validation users for Capture-24 are untouched.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.1 Quantity of the Unlabeled Data Required.
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.
Probably not a substantial claim.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“We plot the mean and standard deviation of the test set mean F1-score obtained after five-fold validation.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.1 Quantity of the Unlabeled Data Required.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“One potential downside could be its effectiveness on multi-sensory datasets due to its heavy reliance on augmentations, as the established sensor augmentations were designed specifically for accelerometry (only).”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 6.3 Advantages and Drawbacks of Self-Supervised Approaches
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“In Fig. 9, we perform five randomized runs for both the pre-training and the linear classification, and plot the difference in the mean of the train set F1-scores.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > CPC Autoencoder
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“We perform three randomized runs, including choosing training subject(s), pre-training, and classification on the target datasets.”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > 5.2.1 Quantity of the Unlabeled Data Required.
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.
Probably not a substantial claim.
The paper relies on this without support.
The ubiquitous nature of body-worn sensors allows for the unobtrusive collection of large amounts of unlabeled data.
“While the collection of annotated data remains a challenge [41], the ubiquitous nature of the sensors themselves allows for the unobtrusive collection of large amounts of unlabeled data, which can subsequently be utilized to learn generic representations.”
The paper relies on this without support.
For many scenarios, only a single wearable sensor is feasible.
“This choice is motivated by the feasibility of only having a single wearable sensor for many scenarios.”
The paper relies on this without support.
The target datasets' activity labels are accurate and reliable.
“All layers of the network are trained using the target dataset labels.”
The paper relies on this without support.
Sensor data from different datasets are comparable after per-channel normalization, so representations transfer across devices and body locations.
“The normalization is performed at a per-channel level such that the train split has zero mean and unit variance.”
The paper relies on this without support.
The Capture-24 dataset is representative of free-living human activity, so representations learned on it are useful for other wearable HAR tasks.
“Capture-24 [4, 18, 81] Wrist 151 -Free living”
The paper relies on this without support.
The 50 Hz sampling rate retains sufficient information for the target activities.
“We downsample the large-scale Capture-24 dataset to 50 Hz as it reduces the computational load for pretraining.Additionally, this is also the lowest native sampling rate in the target datasets.”
The paper relies on this without support.
It may not be possible to recruit a large set of participants willing to wear sensors for long periods, and recording shorter periods may be more practical.
“For data collection, it may not be possible to recruit a large set of participants who are willing to wear sensor(s) for long periods of time.It may be more practical to record data for shorter periods of time in free-living or indeed lab conditions.”
The paper relies on this without support.
Wearables-based movement datasets tend to be imbalanced, with more samples of naturally occurring or easily performed activities than strenuous or rare activities.
“Wearables-based movement datasets tend to be imbalanced, often containing considerably more samples of activities that either occur more naturally or can be performed easily, relative to more physically strenous or rarely occurring activites.”
The paper relies on this without support.
Very limited labeled data scenarios are of vital importance to activity recognition and can arise due to cost and privacy concerns, such as in medical applications.
“In this criterion, we consider a scenario of vital importance to activity recognition, which is when very limited labeled data are available for classification.This situation can arise when both collecting and annotating data maybe difficult due to cost and privacy concerns, such as in medical applications.”
The paper relies on this without support.
The gold-standard for activity recognition generally lies in supervised learning, resulting in the best possible performance.
“The gold-standard for activity recognition generally lies in supervised learning, wherein end-to-end training is performed with the available annotations, resulting in the best possible performance.”
The paper relies on this without support.
Real-world deployment may allow for acquiring a small labeled dataset from users without significant interruptions, such as during device setup.
“Alternatively, real-world deployment may allow for acquiring a small labeled dataset from users without significant interruptions (or during device setup).”
The paper relies on this without support.
The small labeled dataset problem is prevalent in sensor datasets.
“These approaches strike at the heart of the small labeled dataset problem prevalent in sensor datasets, and therefore present the 'pretrain-then-finetune' paradigm shift.”
The paper relies on this without support.
Prior techniques for understanding self-supervised HAR performance have been limited to one view where source and target conditions such as sensor locations and activities are similar.
“Techniques towards understanding their performance has so far been limited to one view -wherein the source and target conditions such as sensor locations and activities are similar.”
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 source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“A combination of convolutional and recurrent layers is proposed in DeepConvLSTM [144], which, even today, remains a strong baseline and is commonly utilized in contemporary HAR works [81,189].”
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 [81]
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
“Haresamudram et al. [81] show that the SimCLR framework [184] and multi-task self-supervised learning [166], outperform other approaches, including BYOL, SimSiam, masked reconstruction, CPC and autoencoders for HAR.”
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 [81]
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
“While end-to-end training is more strongly affected by class imbalance and long-tailed distributions [122], self-supervised learning has been shown to be more robust to such imbalances [81,122].”
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 [81]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“as utilizing self-supervised models from the wrist → waist is typically not as effective as wrist → wrist (where the location is common) or wrist → leg (where activities can have tandem movements between the hands and the legs) [27].”
From Investigating Enhancements to Contrastive Predictive Coding for Human Activity Recognition (2023) · cites this paper as [27]
The cited source says what the claim attributes to it.
In the citing paper
“as [27] showed that the performance is comparable to using the entire train dataset”
From Investigating Enhancements to Contrastive Predictive Coding for Human Activity Recognition (2023) · cites this paper as [27]
The cited source says what the claim attributes to it.
In the citing paper
“As in [27], evaluating across different sensor positions and activities allows us to perform a broad evaluation of the proposed method and understand its strengths and areas of opportunity.”
From Investigating Enhancements to Contrastive Predictive Coding for Human Activity Recognition (2023) · cites this paper as [27]
The cited source says what the claim attributes to it.
In the citing paper
“The stronger performance on the waist-based datasets as well indicates the superior performance of enhanced CPC, as the other methods are not as capable of transferring from wrist → waist (also observed in [27]).”
From Investigating Enhancements to Contrastive Predictive Coding for Human Activity Recognition (2023) · cites this paper as [27]
The cited source says what the claim attributes to it.
In the citing paper
“as matching the frequency is more effective [27].”
From Investigating Enhancements to Contrastive Predictive Coding for Human Activity Recognition (2023) · cites this paper as [27]
The cited source says what the claim attributes to it.
In the citing paper
“as it is beneficial for performance [27].”
From Investigating Enhancements to Contrastive Predictive Coding for Human Activity Recognition (2023) · cites this paper as [27]
The cited source says what the claim attributes to it.
In the citing paper
“Haresamudram et al. [81] conduct a comprehensive comparison of different contrastive and Siamese learning methods for HAR.”
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 [81]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“When some labeled training data are available for the new classes, selfsupervised methods such as SimCLR [184] can be leveraged to fine-tune learned models to effectively recognize new activities as well [81].”
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 [81]
The cited source says what the claim attributes to it.
In the citing paper
“A recent attempt has been made to systematically evaluate the effect of self-supervised techniques by pre-training on Capture-24, providing a good baseline for the performance evaluation in human activity recognition33.”
From Self-supervised learning for human activity recognition using 700,000 person-days of wearable data (2024) · cites this paper as [33]
In the citing paper
“However, recent studies [11,19] question the necessity of extensive pre-training data.They found that many self-supervised methods are robust to data size.For example, Haresamudram et al. [19] show that using approx.10% of the source data for pre-training yields comparable or better results than using an entire original dataset for HAR.”
From How Much Unlabeled Data is Really Needed for Effective Self-Supervised Human Activity Recognition? (2023) · cites this paper as [19]
In the citing paper
“Therefore, self-supervision clearly has the capability of being data efficient, i.e., pre-training only on small amounts of data, with similar downstream performance.”
From How Much Unlabeled Data is Really Needed for Effective Self-Supervised Human Activity Recognition? (2023) · cites this paper as [19]
In the citing paper
“Subsequently, we use participant-based data splitting as detailed in [19] to obtain train-validation-test splits.”
From How Much Unlabeled Data is Really Needed for Effective Self-Supervised Human Activity Recognition? (2023) · cites this paper as [19]
In the citing paper
“it was shown in [16] that matching the sampling rates between pre-training and fine-tuning is important for optimal performance.”
From Towards Learning Discrete Representations via Self-Supervision for Wearables-Based Human Activity Recognition (2024) · cites this paper as [16]
In the citing paper
“prior works such as [16,30] have utilized it as the base for self-supervised pre-training, allowing us to compare our results against those works.”
From Towards Learning Discrete Representations via Self-Supervision for Wearables-Based Human Activity Recognition (2024) · cites this paper as [16]
In the citing paper
“The means and variances from the Capture-24 train split are also applied to all sets from the target datasets for improved performance (as per [16]).”
From Towards Learning Discrete Representations via Self-Supervision for Wearables-Based Human Activity Recognition (2024) · cites this paper as [16]
In the citing paper
“we only pre-train on randomly sampled 10% of the windows from the train split, as it was shown to have comparable performance to using the entire split in [16]”
From Towards Learning Discrete Representations via Self-Supervision for Wearables-Based Human Activity Recognition (2024) · cites this paper as [16]
In the citing paper
“HAR has enabled the field to address one of its most pressing issues: using unlabeled data to build reliable recognition systems with only a few labeled training samples [30].”
From Self-supervised learning: A succinct review (2023) · cites this paper as [30]
In the citing paper
“DNNs enable automatic extraction of features from raw sensor data, eliminating the need for manual feature crafting.”
From Reducing Label Dependency in Human Activity Recognition with Wearables: From Supervised Learning to Novel Weakly Self-Supervised Approaches (2025) · cites this paper as [12]
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 major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.2 Transformation Functions used for SimCLR, SimSiam, and BYOL
“| HHAR | 0.005 | 0.0 | 4096 | 0.0005 | 1e-05 | 53.22 | 2.62 |”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.1.6 SimSiam.
“batch size ∈ {128, 256, 512}.”
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
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.2 Transformation Functions used for SimCLR, SimSiam, and BYOL
“| HHAR | 0.005 | 0.0 | 4096 | 0.0005 | 1e-05 | 53.22 | 2.62 |”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Leg
“BYOL 51.68±2.2736.48±1.7320.36±2.0451.07±1.7145.14±2.3955.84±1.34”
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
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.2 Transformation Functions used for SimCLR, SimSiam, and BYOL
“| SimCLR | lr=0.005, wd=0.0001, batch_size=2048, class. lr=0.0001, class. l2 reg.=0.0005 |”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.1.4 Autoencoder.
“The classification is performed using the learning rates ∈ {1e -4, 5e -4, 1e -5} and L2 regularization ∈ {0.0, 1e -4, 1e -5}.”
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
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.2 Transformation Functions used for SimCLR, SimSiam, and BYOL
“| HHAR | 0.005 | 0.0 | 4096 | 0.0005 | 1e-05 | 53.22 | 2.62 |”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Leg
“SimSiam 54.72±1.3237.63±3.7923.94±0.8550.42±1.6646.04±1.7759.64±4.06”
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
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.2 Transformation Functions used for SimCLR, SimSiam, and BYOL
“| Multi-task | | lr=0.0003, l2 reg.=0.0005, class. lr=0.0003, class. l2 reg.=0 |”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.1.1 Multi-task Self-supervision.
“For classification, the parameters include learning rates ∈ {1e -4, 3e -4, 5e -4} and L2 reguralization ∈ {1e -4, 3e -4, 5e -4}.”
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
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.2 Transformation Functions used for SimCLR, SimSiam, and BYOL
“| Myogym | 0.0005 0.00001 0.0005 | 0.0001 | 39.89 | 0.98 | | |”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > Leg
“MLP classifier Multi-task self.sup 57.51 ±1.9 42.31 ±2.37 23.35 ±0.66”
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
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.1.4 Autoencoder.
“As in [72], the convolutional encoder for this method is identical to the one used in Multi-task self-supervision [62] and comprises of three 1D convolutional blocks of 23, 64, and 96 filters, respectively.The corresponding kernel sizes are 24, 16, and 8”
§ Assessing the State of Self-Supervised Human Activity Recognition Using Wearables > A.1.1 Multi-task Self-supervision.
“The encoder architecture is identical to the original paper [62], which contained three 1D convolutional layers, having 32, 64 and 96 filters, with a kernel size of 24, 18 and 8, respectively.”
Citations
Cites
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
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
(2008) · Journal of Machine Learning Research · Cited by 33,051
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
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Ioffe, S., & Szegedy, C. (2015) · arXiv · doi:10.48550/arxiv.1502.03167 · Cited by 27,818
Distilling the Knowledge in a Neural Network
Hinton, G., Vinyals, O., & Dean, J. (2015) · arXiv · doi:10.48550/arxiv.1503.02531 · Cited by 14,204
Momentum Contrast for Unsupervised Visual Representation Learning
He, K., Fan, H., Wu, Y., Xie, S., & Girshick, R. (2020) · 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr42600.2020.00975 · Cited by 13,347 · cited 1×
“Over the last few years, methods such as SimCLR [6], MoCo [28], SimSiam [7], NNCLR [15], and Barlow Twins [83] have shown increasing effectiveness for representation learning and constitute of the state-of-the-art in computer vision.”
MethodEmpirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling
Chung, J., Gulcehre, C., Cho, K., & Bengio, Y. (2014) · arXiv · doi:10.48550/arxiv.1412.3555 · Cited by 11,008
A simple framework for contrastive learning of visual representations
Chen, T., Kornblith, S., Norouzi, M., & Hinton, G. (2020) · arXiv · doi:10.48550/arxiv.2002.05709 · Cited by 10,857
(2012) · Cited by 8,009
Representation Learning with Contrastive Predictive Coding
Oord, A. V. D., Li, Y., & Vinyals, O. (2018) · arXiv · doi:10.48550/arxiv.1807.03748 · Cited by 6,242
Evaluation: from precision, recall and F-measure to ROC, informedness, markedness and correlation
Powers, D. M. W. (2020) · arXiv · doi:10.48550/arxiv.2010.16061 · Cited by 5,440
Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps
Simonyan, K., Vedaldi, A., & Zisserman, A. (2013) · arXiv · doi:10.48550/arxiv.1312.6034 · Cited by 5,370
Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning
Grill, J.-B., Strub, F., Altché, F., Tallec, C., Richemond, P. H., Buchatskaya, E., Doersch, C., Pires, B. A., Guo, Z. D., Azar, M. G., et al. (2020) · arXiv · doi:10.48550/arxiv.2006.07733 · Cited by 3,846 · cited 5×
“Finally, we also evaluate BYOL [21], which utilizes on two neural networks, namely the online and target networks, to interact and learn from one another.”
BackgroundMethodInfluentialExploring Simple Siamese Representation Learning
Chen, X., & He, K. (2021) · 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr46437.2021.01549 · Cited by 3,725 · cited 6×
“SimSiam [7] produces surprisingly effective representations with a simple siamese setup.”
BackgroundMethodResultInfluentialDeep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition
Ordóñez, F., & Roggen, D. (2016) · Sensors · doi:10.3390/s16010115 · Cited by 2,761
Unsupervised Representation Learning by Predicting Image Rotations
Gidaris, S., Singh, P., & Komodakis, N. (2018) · arXiv · doi:10.48550/arxiv.1803.07728 · Cited by 2,477 · cited 1×
“These methods have shown great promise across many domains, including computer vision [22], natural language processing [16, 54, 55], and speech processing [11–13].”
MethodBag of Tricks for Image Classification with Convolutional Neural Networks
He, T., Zhang, Z., Zhang, H., Zhang, Z., Xie, J., & Li, M. (2019) · 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2019.00065 · Cited by 1,680 · cited 1×
“This practice of applying normalization using the source dataset means and variances is utilized for transfer learning [32].”
MethodBarlow Twins: Self-Supervised Learning via Redundancy ReductionPreprint
Žbontar, J., Jing, L., Misra, I., LeCun, Y., & Deny, S. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.03230 · Cited by 1,506
Introducing a New Benchmarked Dataset for Activity Monitoring
Reiss, A., & Stricker, D. (2012) · 2012 16th International Symposium on Wearable Computers · doi:10.1109/iswc.2012.13 · Cited by 1,456 · cited 14×
“2 BACKGROUND AND MOTIVATION Human activity recognition (HAR) is typically a five step process, as outlined in the activity recognition chain [57] – data collection, pre-processing, segmentation, feature extraction, and classification.”
BackgroundMethodInfluentialwav2vec: Unsupervised Pre-Training for Speech Recognition
Schneider, S., Baevski, A., Collobert, R., & Auli, M. (2019) · Interspeech 2019 · doi:10.21437/interspeech.2019-1873 · Cited by 1,290
A Transformer-based Framework for Multivariate Time Series Representation Learning
Zerveas, G., Jayaraman, S., Patel, D., Bhamidipaty, A., & Eickhoff, C. (2021) · Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining · doi:10.1145/3447548.3467401 · Cited by 1,278
Doherty, A., Jackson, D., Hammerla, N., Plötz, T., Olivier, P., Granat, M. H., White, T., van Hees, V. T., Trenell, M. I., Owen, C. G., et al. (2017) · PLOS ONE · doi:10.1371/journal.pone.0169649 · Cited by 1,250 · cited 1×
“Techniques that require smaller training sets are more advantageous as they result in reduced computational costs, a factor that is very important for wearables.”
BackgroundSelf-citationUm, T. T., Pfister, F. M. J., Pichler, D., Endo, S., Lang, M., Hirche, S., Fietzek, U., & Kulić, D. (2017) · Proceedings of the 19th ACM International Conference on Multimodal Interaction · doi:10.1145/3136755.3136817 · Cited by 729
Similarity of Neural Network Representations Revisited
Kornblith, S., Norouzi, M., Lee, H., & Hinton, G. (2019) · arXiv · doi:10.48550/arxiv.1905.00414 · Cited by 724
Stisen, A., Blunck, H., Bhattacharya, S., Prentow, T. S., Kjærgaard, M. B., Dey, A., Sonne, T., & Jensen, M. M. (2015) · Proceedings of the 13th ACM Conference on Embedded Networked Sensor Systems · doi:10.1145/2809695.2809718 · Cited by 714 · cited 8×
“Further, we observe how having the same target location (i.e., the wrist) can result in increased performance over all supervised baselines, specially in the case of HHAR and Myogym.”
MethodInfluentialWearable Assistant for Parkinson’s Disease Patients With the Freezing of Gait Symptom
Bachlin, M., Plotnik, M., Roggen, D., Maidan, I., Hausdorff, J. M., Giladi, N., & Troster, G. (2010) · IEEE Transactions on Information Technology in Biomedicine · doi:10.1109/titb.2009.2036165 · Cited by 693 · cited 8×
“For datasets such as Myogym, Wetlab, Daphnet FoG etc., which contain activities that are at least not partially covered in the Capture-24 dataset (see Tab.”
MethodInfluentialRevisiting Self-Supervised Visual Representation Learning
Kolesnikov, A., Zhai, X., & Beyer, L. (2019) · 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2019.00202 · Cited by 675 · cited 2×
“At the same time, larger representation sizes can benefit from the availability of more training data [38].”
BackgroundUm, T. T. (n.d.) · mediaTUM – the media and publications repository of the Technical University Munich (Technical University Munich) · Cited by 602
mHealthDroid: A Novel Framework for Agile Development of Mobile Health Applications
Banos, O., Garcia, R., Holgado-Terriza, J. A., Damas, M., Pomares, H., Rojas, I., Saez, A., & Villalonga, C. (2014) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-13105-4_14 · Cited by 583 · cited 12×
“In comparison, transferring from the wrist to leg results in comparable performance (Daphnet FoG and PAMAP2, against the Conv. classifier), or gains (MHEALTH).”
MethodInfluentialGLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding
Wang, A., Singh, A., Michael, J., Hill, F., Levy, O., & Bowman, S. R. (2018) · arXiv · doi:10.48550/arxiv.1804.07461 · Cited by 562
USC-HAD: a daily activity dataset for ubiquitous activity recognition using wearable sensors
Zhang, M., & Sawchuk, A. A. (2012) · Proceedings of the 2012 ACM Conference on Ubiquitous Computing · doi:10.1145/2370216.2370438 · Cited by 554 · cited 8×
“Similarly, it is interesting to see that the Autoencoder shows the best performance using just ∼ 410 windows on some datasets (such as Myogym, USC-HAD, Mobiact) etc.”
MethodInfluentialWith a Little Help from My Friends: Nearest-Neighbor Contrastive Learning of Visual Representations
Dwibedi, D., Aytar, Y., Tompson, J., Sermanet, P., & Zisserman, A. (2021) · arXiv · doi:10.48550/arxiv.2104.14548 · Cited by 405
With a Little Help from My Friends: Nearest-Neighbor Contrastive Learning of Visual Representations
Dwibedi, D., Aytar, Y., Tompson, J., Sermanet, P., & Zisserman, A. (2021) · 2021 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00945 · Cited by 404 · cited 1×
“Over the last few years, methods such as SimCLR [6], MoCo [28], SimSiam [7], NNCLR [15], and Barlow Twins [83] have shown increasing effectiveness for representation learning and constitute of the state-of-the-art in computer vision.”
MethodExploring Simple Siamese Representation Learning
Chen, X., & He, K. (2020) · arXiv · doi:10.48550/arxiv.2011.10566 · Cited by 381
Mockingjay: Unsupervised Speech Representation Learning with Deep Bidirectional Transformer Encoders
Liu, A. T., Yang, S.-W., Chi, P.-H., Hsu, P.-C., & Lee, H.-Y. (2020) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp40776.2020.9054458 · Cited by 352 · cited 1×
“For example, masked prediction of log-mel spectrograms was studied in [80] for speech recognition, with the concept being extended in [43, 44, 86].”
MethodMulti-task Self-Supervised Learning for Human Activity Detection
Saeed, A., Ozcelebi, T., & Lukkien, J. (2019) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3328932 · Cited by 346 · cited 7×
“3 and 4, as it has been shown to be very effective compared to grid searches [3], while also not consuming extensive computational resources.”
BackgroundMethodInfluentialAnalyzing features for activity recognition
Huỳnh, T., & Schiele, B. (2005) · Proceedings of the 2005 joint conference on Smart objects and ambient intelligence: innovative context-aware services: usages and technologies · doi:10.1145/1107548.1107591 · Cited by 340 · cited 1×
“They follow the ‘pretrain-then-finetune’ training paradigm (see Fig.”
BackgroundFeature Learning for Activity Recognition in Ubiquitous Computing
Plötz, T., Hammerla, N., & Olivier, P. (2011) · Newcastle University · doi:10.25405/ncl.34175319 · Cited by 337
Self-citationRaghu, M., Gilmer, J., Yosinski, J., & Sohl-Dickstein, J. (2017) · arXiv · doi:10.48550/arxiv.1706.05806 · Cited by 335
TERA: Self-Supervised Learning of Transformer Encoder Representation for Speech
Liu, A. T., Li, S.-W. D., & Lee, H.-Y. (2021) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2021.3095662 · Cited by 333 · cited 1×
“For example, masked prediction of log-mel spectrograms was studied in [80] for speech recognition, with the concept being extended in [43, 44, 86].”
MethodRethinking the Value of Labels for Improving Class-Imbalanced LearningPreprint
Yang, Y., & Xu, Z. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.07529 · Cited by 320 · cited 2×
“As in [43, 74] we first define the imbalance ratio ρ as the ratio of the rarest class to the most frequent class.”
BackgroundWilletts, M., Hollowell, S., Aslett, L., Holmes, C., & Doherty, A. (2018) · Scientific Reports · doi:10.1038/s41598-018-26174-1 · Cited by 250
Lotta, L. A., Wittemans, L. B. L., Zuber, V., Stewart, I. D., Sharp, S. J., Luan, J., Day, F. R., Li, C., Bowker, N., Cai, L., et al. (2018) · JAMA · doi:10.1001/jama.2018.19329 · Cited by 229 · cited 2×
“1 and detailed below.”
BackgroundPD Disease State Assessment in Naturalistic Environments Using Deep Learning
Hammerla, N., Fisher, J., Andras, P., Rochester, L., Walker, R., & Ploetz, T. (2015) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v29i1.9484 · Cited by 179
Optimising sampling rates for accelerometer-based human activity recognition
Khan, A., Hammerla, N., Mellor, S., & Plötz, T. (2016) · Pattern Recognition Letters · doi:10.1016/j.patrec.2016.01.001 · Cited by 171 · cited 1×
“We established that pre-training on a large-scale dataset collected at the wrist has a positive impact on the target activity recognition, even when the sensor locations are different or if the activities are not covered during pre-training.”
BackgroundSelf-citationUnsupervised representation learning for time series with temporal neighborhood coding
Tonekaboni, S., Eytan, D., & Goldenberg, A. (2021) · arXiv · doi:10.48550/arxiv.2106.00750 · Cited by 170
Protecting sensory data against sensitive inferences
Malekzadeh, M., Clegg, R. G., Cavallaro, A., & Haddadi, H. (2018) · Spiral (Imperial College London) · Cited by 166
Generative Pre-Training for Speech with Autoregressive Predictive Coding
Chung, Y.-A., & Glass, J. (2020) · ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp40776.2020.9054438 · Cited by 157
Protecting Sensory Data against Sensitive Inferences
Malekzadeh, M., Clegg, R. G., Cavallaro, A., & Haddadi, H. (2018) · Proceedings of the 1st Workshop on Privacy by Design in Distributed Systems · doi:10.1145/3195258.3195260 · Cited by 154 · cited 2×
“15 for reference) with Wetlab, Mobiact, MHEALTH and Myogym having the most severe class compositions.”
MethodKwon, 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
Self-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 · cited 5×
“The solved tasks need to be replaced with more challenging and relevant ones, addressing discovered flaws in the established experimental setups [17].”
BackgroundMethodInfluentialSelf-citationMasked reconstruction based self-supervision for human activity recognition
Haresamudram, H., Beedu, A., Agrawal, V., Grady, P. L., Essa, I., Hoffman, J., & Plötz, T. (2020) · Proceedings of the 2020 International Symposium on Wearable Computers · doi:10.1145/3410531.3414306 · Cited by 146 · cited 5×
“…and BYOL we randomly sample 20 sets of hyperparameter combinations during pre-training (the parameter spaces of which are detailed below), whereas for Multi-task self-supervision, we perform a grid search over the learning rates and L2 regularizations as the possible combinations are fewer than 20.”
BackgroundMethodInfluentialSelf-citationHammerla, N. Y., Kirkham, R., Andras, P., & Ploetz, T. (2013) · Proceedings of the 2013 International Symposium on Wearable Computers · doi:10.1145/2493988.2494353 · Cited by 138 · cited 1×
“[32, 55]; (ii) distribution-based features, which comprise the state-of-the-art for feature extraction, involving the inverse of the cumulative distribution function [24]; and (iii) learned features, consisting of dimensionality reduction techniques, supervised and unsupervised (incl.”
Time Series Change Point Detection with Self-Supervised Contrastive Predictive Coding
Deldari, S., Smith, D. V., Xue, H., & Salim, F. D. (2021) · Proceedings of the Web Conference 2021 · doi:10.1145/3442381.3449903 · Cited by 136 · cited 1×
“Such a system facilitates more nuanced interpretations of the results while rejecting hill climbing prevalent in leaderboards.”
BackgroundO’Connor, L., Brage, S., Griffin, S. J., Wareham, N. J., & Forouhi, N. G. (2015) · British Journal of Nutrition · doi:10.1017/s000711451500269x · Cited by 125 · cited 4×
“Effective performance across differing sensor locations allows for leveraging large scale unlabeled datasets– which are typically collected at the wrist (such as Capture-24 [4, 18, 81], Fenland [46, 51], and Biobank [14]), to jumpstart recognition for rarer leg-based activities such as classifying freeze of gait.”
BackgroundMethodInfluentialTesting Self-Report Time-Use Diaries against Objective Instruments in Real Time
Gershuny, J., Harms, T., Doherty, A., Thomas, E., Milton, K., Kelly, P., & Foster, C. (2020) · Sociological Methodology · doi:10.1177/0081175019884591 · Cited by 124
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
Federated Self-Supervised Learning of Multisensor Representations for Embedded Intelligence
Saeed, A., Salim, F. D., Ozcelebi, T., & Lukkien, J. (2021) · IEEE Internet of Things Journal · doi:10.1109/jiot.2020.3009358 · Cited by 106 · cited 1×
“Other self-supervised approaches in activity recognition include signal correspondence learning with the wavelet transform in a federated learning framework [59], utilizing eight auxiliary tasks for learning high-level features [60], contrastive learning for activity recognition in healthcare-based applications [66].”
BackgroundDeep Learning for Human Activity Recognition in Mobile Computing
Plotz, T., & Guan, Y. (2018) · Computer · doi:10.1109/mc.2018.2381112 · Cited by 99 · cited 1×
“Methods, which are effective without requiring many users or hours of data are preferable [51].”
BackgroundSelf-citationFederated Self-Supervised Learning of Multisensor Representations for Embedded Intelligence
Saeed, A., Salim, F. D., Özçelebi, T., & Lukkien, J. J. (2024) · Figshare · Cited by 92
Let's (not) stick together: pairwise similarity biases cross-validation in activity recognition
Hammerla, N. Y., & Plötz, T. (2015) · Proceedings of the 2015 ACM International Joint Conference on Pervasive and Ubiquitous Computing · doi:10.1145/2750858.2807551 · Cited by 91 · cited 1×
“Overall, every participant appears exactly once as part of the validation and test splits and pairwise correlations of analysis windows are effectively eliminated through the user-based splits [26].”
BackgroundSelf-citationHuman Daily Activity and Fall Recognition Using a Smartphone’s Acceleration Sensor
Chatzaki, C., Pediaditis, M., Vavoulas, G., & Tsiknakis, M. (2017) · Communications in Computer and Information Science · doi:10.1007/978-3-319-62704-5_7 · Cited by 91
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 · cited 24×
“Similar to the original papers, the learning rate for classifiers for CPC, Masked reconstruction, and Autoencoder [25] reduces by a factor of 0.8 every 10 epochs.”
BackgroundMethodResultInfluentialSelf-citationSelf-supervised Learning is More Robust to Dataset Imbalance
Liu, H., HaoChen, J. Z., Gaidon, A., & Ma, T. (2021) · arXiv · doi:10.48550/arxiv.2110.05025 · Cited by 84
Unsupervised Pre-Training of Bidirectional Speech Encoders via Masked Reconstruction
Wang, W., Tang, Q., & Livescu, K. (2020) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp40776.2020.9053541 · Cited by 83 · cited 1×
“For example, masked prediction of log-mel spectrograms was studied in [80] for speech recognition, with the concept being extended in [43, 44, 86].”
MethodExploring Contrastive Learning in Human Activity Recognition for Healthcare
Tang, C. I., Perez-Pozuelo, I., Spathis, D., & Mascolo, C. (2020) · arXiv · doi:10.48550/arxiv.2011.11542 · Cited by 67
On the Intrinsic Dimensionality of Image Representations
Gong, S., Boddeti, V. N., & Jain, A. K. (2019) · 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2019.00411 · Cited by 59 · cited 2×
“Implicit dimensionality of the representations – captures the true capacity and redundancies in the representations learned by self-supervision [23].”
BackgroundMethodImproved Speech Representations with Multi-Target Autoregressive Predictive CodingPreprint
Chung, Y.-A., & Glass, J. R. (2020) · Annual Meeting of the Association for Computational Linguistics · doi:10.48550/arxiv.2004.05274 · Cited by 58
The GEM Benchmark: Natural Language Generation, its Evaluation and Metrics
Gehrmann, S., Adewumi, T., Aggarwal, K., Ammanamanchi, P. S., Anuoluwapo, A., Bosselut, A., Chandu, K. R., Clinciu, M., Das, D., Dhole, K. D., et al. (2021) · arXiv · doi:10.48550/arxiv.2102.01672 · Cited by 49
MyoGym: introducing an open gym data set for activity recognition collected using myo armband
Koskimäki, H., Siirtola, P., & Röning, J. (2017) · Proceedings of the 2017 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of the 2017 ACM International Symposium on Wearable Computers · doi:10.1145/3123024.3124400 · Cited by 46 · cited 18×
“Similarly, it is interesting to see that the Autoencoder shows the best performance using just ∼ 410 windows on some datasets (such as Myogym, USC-HAD, Mobiact) etc.”
MethodInfluentialWearables in the wet lab: a laboratory system for capturing and guiding experiments
Scholl, P. M., Wille, M., & Van Laerhoven, K. (2015) · Proceedings of the 2015 ACM International Joint Conference on Pervasive and Ubiquitous Computing · doi:10.1145/2750858.2807547 · Cited by 45 · cited 11×
“A similar trend is observed for the lab-based gestures in Wetlab as well, with > 10% increase shown by CPC, Autoencoder, SimSiam, and BYOL.”
MethodInfluentialAn Unsupervised Autoregressive Model for Speech Representation LearningPreprint
Chung, Y.-A., Hsu, W.-N., Tang, H., et al. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1904.03240 · Cited by 45
Extending and Analyzing Self-Supervised Learning Across Domains
Wallace, B., & Hariharan, B. (2020) · European Conference on Computer Vision · doi:10.48550/arxiv.2004.11992 · Cited by 45
Time-Series Representation Learning via Temporal and Contextual ContrastingPreprint
Eldele, E., Ragab, M., Chen, Z., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.14112 · Cited by 36
Extending and Analyzing Self-supervised Learning Across Domains
Wallace, B., & Hariharan, B. (2020) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-58574-7_43 · Cited by 35
PlÖtz, T. (2022) · ACM Computing Surveys · doi:10.1145/3459666 · Cited by 34
Self-citationSense and Learn: Self-supervision for omnipresent sensors
Saeed, A., Ungureanu, V., & Gfeller, B. (2021) · Machine Learning with Applications · doi:10.1016/j.mlwa.2021.100152 · Cited by 34
Unsupervised Pre-Training Of Bidirectional Speech Encoders Via Masked ReconstructionPreprint
Wang, W., Tang, Q., & Livescu, K. (2020) · arXiv (Cornell University) · doi:10.17023/p6jq-6c54 · Cited by 26
Kwon, H., Tong, C., Haresamudram, H., Gao, Y., Abowd, G. D., Lane, N. D., & Ploetz, T. (2020) · arXiv · doi:10.48550/arxiv.2006.05675 · Cited by 14
Self-citationMusiCoder: A Universal Music-Acoustic Encoder Based on Transformer
Zhao, Y., & Guo, J. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-67832-6_34 · Cited by 13
Deriving Effective Human Activity Recognition Systems through Objective Task Complexity Assessment
Hiremath, S. K., & Plötz, T. (2020) · Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · doi:10.1145/3432227 · Cited by 11 · cited 1×
“This is compounded by the necessity to enlist healthcare professionals for annotation, which may incur high costs [31].”
BackgroundSelf-citationGenerative Pre-Training for Speech with Autoregressive Predictive CodingPreprint
Chung, Y.-A., & Glass, J. R. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1910.12607 · Cited by 10
Capture-24: Activity tracker dataset for human activity recognition
Chan Chang, S., & Doherty, A. (2021) · University of Oxford · doi:10.5287/bodleian:nzvqj2ojw · Cited by 8
Evaluating Contrastive Learning on Wearable Timeseries for Downstream Clinical OutcomesPreprint
Shah, K., Spathis, D., Tang, C. I., & Cecilia Mascolo, C. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2111.07089 · Cited by 4
ColloSSL: Collaborative Self-Supervised Learning for Human Activity RecognitionPreprint
Jain, Y., Tang, C. I., Min, C., Kawsar, F., & Mathur, A. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2202.00758 · Cited by 3
Self-supervised Deep Learning for Reading Activity ClassificationPreprint
Islam, M. R., Sakamoto, S., Yamada, Y., Vargo, A. W., Iwata, M., Iwamura, M., & Kise, K. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2012.03598
What Makes Good Contrastive Learning on Small-Scale Wearable-based Tasks?Preprint
Qian, H., Tian, T., & Miao, C. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2202.05998
Doherty, A., Jackson, D., Hammerla, N., et al. (2017) · Newcastle University ePrints (Newcastle Univesity) · doi:10.25405/ncl.34308291
Wearable Assistant for Parkinson's Disease Patients With the Freezing of Gait Symptom
Bächlin, M., Plotnik, M., Roggen, D., et al. (2010) · Newcastle University ePrints (Newcastle Univesity) · doi:10.25405/ncl.34198923
Let’s (not) Stick Together: Pairwise Similarity Biases Cross-Validation in Activity Recognition
Hammerla, N., & Ploetz, T. (2015) · Newcastle University ePrints (Newcastle Univesity) · doi:10.25405/ncl.34259634
SelfHAR: Improving Human Activity Recognition through Self-training with Unlabeled Data
Tang, C. I., Perez-Pozuelo, I., Spathis, D., et al. (2021) · arXiv · doi:10.48550/arxiv.2102.06073
wav2vec: Unsupervised Pre-training for Speech Recognition
Schneider, S., Baevski, A., Collobert, R., et al. (2019) · arXiv · doi:10.48550/arxiv.1904.05862
Research works printed but not linked: 4
- 10.5555/2886521.2886562 · DOI
- Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. 2014. Dropout: a simple way to prevent neural networks from overfitting. The journal of machine learning research 15, 1 (2014), 1929--1958.
- Aaron Van den Oord, Yazhe Li, and Oriol Vinyals. 2018. Representation learning with contrastive predictive coding. arXiv e-prints (2018), arXiv-1807.
- Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. Advances in neural information processing systems 30 (2017).
Cited by
Sorted by citations
Self-supervised learning: A succinct review
Rani, V., Nabi, S. T., Kumar, M., Mittal, A., & Kumar, K. (2023) · Archives of Computational Methods in Engineering · doi:10.1007/s11831-023-09884-2 · Cited by 295
“HAR has enabled the field to address one of its most pressing issues: using unlabeled data to build reliable recognition systems with only a few labeled training samples [30].”
BackgroundSelf-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
“…(SSL) methods to leverage unlabelled data in a similar manner as language models (Mikolov et al., 2013; Devlin et al., 2018; Radford et al., 2018, 2019) and vision models (Doersch et al., 2015; Zhang et al., 2016; Noroozi and Favaro, 2016; Wei et al., 2018; He et al., 2020; Chen et al., 2020).”
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].”
BackgroundChan, S., Hang, Y., Tong, C., Acquah, A., Schonfeldt, A., Gershuny, J., & Doherty, A. (2024) · Scientific Data · doi:10.1038/s41597-024-03960-3 · Cited by 61
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
“Leng et al. [43] piloted the concept of IMUGPT that combines large language models (LLMs), motion synthesis methods, and signal processing techniques to generate virtual IMU data with no manual effort.”
Self-citation
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
- 2022-09-06
- Volume
- 6
- Issue
- 3
- Pages
- 1-47
- ISSN
- 2474-9567 (ISSN-L)
- OpenAlex
- W4294891660
- Semantic Scholar
- 9ce05c848b1a0c05a4ea44b442386c496d2a9a36
- Cited by
- 107
- References
- 49
- Citations by year
- 2026: 15 · 2025: 37 · 2024: 37 · 2023: 17 · 2022: 1
- FWCI
- 7.53
- Type
- journal article
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- National Science Foundation (2112633)
- 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
- Health Sciences › Medicine › Public Health, Environmental and Occupational Health › Physical Activity and Health
- Keywords
- self-supervised learning, human activity recognition, wearable sensors, masked reconstruction, contrastive predictive coding, SimCLR, cross-domain robustness, unlabeled data, representation learning, performance evaluation framework
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 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-11 (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
