REVIEW 4 major objections 5 minor 1 cited by
Subject Invariant Contrastive Learning for Human Activity Recognition
T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash
Pith's one-line read The paper claims that contrastive HAR training embeds subject-specific cues that hurt generalization, and that reweighting same-subject negatives in the loss suppresses those cues and improves accuracy on unseen subjects.
desk verdict Modest but plausible extension of the authors' own taxonomy reweighting to subject labels; consistent gains across three datasets, but the load-bearing weighting term QSi is never defined as an equation, which needs fixing before the result is fully trustworthy. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the SICL loss, an altered contrastive denominator. With $z_i$ the anchor embedding and $\tau$ the temperature, the loss is $L_{\mathrm{SICL}} = -\sum_i \log\left( \frac{\exp(z_i \cdot z_j / \tau)}{D_i} \right)$, where $D_i = Q_{S_i} \sum_{s \in S(i)} \exp(z_i \cdot z_s / \tau) + \sum_{k \notin S(i)} \exp(z_i \cdot z_k / \tau)$. Here $S(i)$ is the set of negatives sharing the anchor's subject; $Q_{S_i}$ is computed batch-wise by exponentiating cosine similarities of same-subject negatives, summing them, and dividing by their average. Multiplying the same-subject sum by $Q_{S_i}$ inflates that part of the denominator, so minimizing the loss pushes anchors away from other samples of the same subject. Subject labels are used only during training, so the learned encoder remains usable without labels at test time.
What would settle it
Run SICL and SimCLR on the same subject-disjoint splits, then rerun SICL with the same-subject weight set to 1 everywhere so the loss reduces to the standard contrastive denominator; if held-out-subject accuracy does not change, the reported gains do not come from the subject weighting. A complementary check is to use data where each subject performs only one activity, so subject identity and activity are perfectly confounded; there, suppressing subject cues should hurt SICL relative to SimCLR.
Extended reading notes
Core claim
The paper's central claim is that the distribution shift induced by subject variability is a primary reason contrastive HAR representations fail on unseen subjects. The authors first show evidence for the shift: on the DARai dataset, spectral entropy and dominant frequency differ significantly across subjects in one-way ANOVAs, seen-subject accuracy exceeds unseen-subject accuracy for SimCLR, and intra-subject pairs have higher cosine similarity than all pairs after standard contrastive training. SICL then changes the contrastive denominator by splitting negatives into same-subject and other-subject sets and multiplying the same-subject terms by a batch-wise weight $Q_{S_i}$ built from exponentiated cosine similarities of same-subject negatives. This raises the penalty for leaving embeddings of the same person close together, so the encoder must separate samples by activity rather than by subject identity. Across UTD-MHAD, MMAct, and DARai, the paper reports consistent top-1 accuracy gains over SimCLR, hard-negative contrastive loss, Barlow Twins, and VICReg, and also over the CMC and CMC-CMKM multimodal frameworks and the supervised contrastive loss.
Load-bearing premise
The load-bearing premise is that reweighting negative pairs from the same subject pushes those embeddings apart without harming the separation between different activities, and that subject identity is the dominant shift in the data.
Editorial extensions
If this is right
- Adding SICL to SimCLR, Barlow Twins, or VICReg raises linear-evaluation top-1 accuracy on held-out subjects across UTD-MHAD, MMAct, and DARai.
- SICL extends to multimodal contrastive frameworks: it improves CMC and CMC-CMKM on inertial-plus-skeleton and inertial-plus-foot-pressure inputs.
- Using SICL as pretraining before cross-entropy fine-tuning beats no pretraining, SimCLR pretraining, and hard-negative pretraining in the reported tables.
- Applying subject reweighting to the supervised contrastive loss (SI-SupCon) improves over both SupCon and SupHCL, so the mechanism transfers to label-aware training.
- Trained representations become more subject-invariant: the cosine-similarity distribution of intra-subject pairs moves closer to the all-pairs distribution.
Reading between the lines
- The same reweighting recipe should transfer to other sensor domains with a known recording source, such as per-patient medical time series or per-device wearables, where source identity creates a similar distribution shift.
- Because SICL needs only subject labels, it could be applied in federated or privacy-preserving settings where each client is one subject and activity labels are unavailable.
- A closed-form definition of $Q_{S_i}$ would let practitioners test sensitivity to the weighting scheme, and making this weight a learned function of similarity could adapt suppression strength per subject.
- SICL's gain could serve as a diagnostic: if adding the term does not help on a given dataset, the dominant distribution shift in that dataset is probably not subject identity.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This paper proposes Subject-Invariant Contrastive Learning (SICL), a modification of the contrastive loss that re-weights negative pairs drawn from the same subject. The goal is to reduce subject-specific information in learned embeddings and improve generalization to unseen subjects in human activity recognition. The loss is applied to unimodal and multimodal contrastive frameworks, as well as supervised contrastive learning. The authors evaluate on UTD-MHAD, MMAct, and DARai, reporting consistent improvements over baselines in linear evaluation, fine-tuning, and supervised settings.
Significance. The paper addresses an important and practical problem: subject variability in HAR causes domain shift, and standard contrastive learning does not account for it. The proposed loss is conceptually simple and can be plugged into several existing methods (SimCLR, Barlow Twins, VICReg, CMC, CMC-CMKM, SupCon). The authors provide a public code repository and use openly available datasets, and they include a motivating analysis (ANOVA, cosine similarity distributions) showing that subject shift exists. If the mechanism is made precise and the empirical results are verified with proper statistical support, the method would be a useful contribution to the HAR community. However, at present the key weighting function is not formally defined and the experimental evaluation lacks error bars, which tempers the strength of the empirical claims.
major comments (4)
- [Section 4.1, Eq. (3)] The weighting function QSi, which is the central new mechanism of SICL, is never defined as a formula. The prose description in Section 4.1 ('normalizing the exponentiated cosine similarities of same-subject negatives, summing these values, and then dividing by their average') admits multiple implementations: normalization could be by sum, by max, or by L2 norm; it could be computed per anchor or globally per batch; and the 'average' division could produce a scalar multiplier or a set of per-pair weights. Because QSi (and its supervised analogue QSij in Eqs. (5) and (7)) multiplies the entire same-subject negative sum in Eq. (3), these choices change the optimization landscape and could even alter the sign or relative magnitude of the penalty. The reported gains in Tables 1-4 are therefore not attributable to a specific re-weighting scheme. Please provide the exact functional form of QSi and QSij, or a reference to a specific equation in [17] or [18] that defines them.
- [Section 5.1, Tables 1-4] The experimental protocol for the unseen-subject evaluation is underspecified. The paper does not state how subjects are partitioned into training and test sets for each dataset (e.g., the number of training subjects, whether multiple splits are used, or whether results are averaged across seeds). This is critical because some reported improvements are small: for example, in Table 1 on DARai inertial data, SimCLR+SICL (0.1610) and SICL (0.1615) differ by only 0.0005, and the absolute gain over SimCLR (0.1498) is about 0.011. Without error bars or repeated runs, the claim of 'consistent improvements' across all cells is not statistically supported. Please provide variance estimates (e.g., standard deviations over at least three seeds) and a precise description of the subject split.
- [Section 4.1, Eq. (1)-(3)] In the self-supervised contrastive setup of Eq. (1), the negative set A(i) contains all other instances in the batch, including time windows from the same subject performing the same activity as the anchor. Reweighting all same-subject negatives via QSi in Eq. (3) therefore also pushes apart representations of identical activities within a subject. The paper does not analyze this effect, nor does it provide a control experiment that differentiates between suppressing subject identity and suppressing same-activity structure. Figure 3 only shows cosine similarities of all pairs versus intra-subject pairs; it does not condition on activity labels. Since the method's stated goal is to remove subject-specific cues while retaining activity information, an experiment that varies the composition of S(i) (e.g., excluding same-activity negatives) would be necessary to support that causal claim.
- [Section 5.3.1, Figure 3] The paper claims that SICL learns subject-invariant representations, but no direct measure of subject invariance is reported. The only evidence is downstream activity classification accuracy and the aggregate cosine similarity distributions in Figure 3. A more direct test would be to train a classifier to predict subject identity from the learned embeddings: if the method indeed suppresses subject-specific cues, subject-classification accuracy should drop substantially relative to SimCLR while activity accuracy is maintained. This experiment would strengthen the central claim and is a natural complement to the ANOVA analysis in Section 3.
minor comments (5)
- [Figure 2 caption] The caption of Figure 2 is incomplete: it ends with 'In the seen subject setting.' followed by nothing, and the sentence about the t-SNE plot is cut off. Please fix the caption.
- [Introduction, paragraph 3] The sentence 'This framework has been adapted for sensor-based human activity recognition by employing signal augmentations to form positive pairs during contrastive loss computation [9]. Negative samples are again drawn uniformly from the data. Recent methods have suggested more informative negative sampling strategies... However, these methods do not consider distribution shifts in HAR data during the sampling of negatives.' is missing a verb or subject in the first clause; the phrase 'without considering their informativeness for learning discriminative representations' in the previous sentence is unclear. Consider rewriting for clarity.
- [Section 5.1] For DARai, the paper states 'We use a subset of the dataset consisting of 20 subjects' but does not specify which subjects or how the subset is selected. Please provide this information to ensure reproducibility.
- [Section 5.2] The sentence 'We use the same temperature parameters as in [11]' is vague. Please list the temperature values used for each dataset and modality, or provide a reference to the specific configuration.
- [Abstract and Section 6] The abstract claims improvement 'up to 11%', but the largest relative improvement visible in Table 1 is approximately 10% (Barlow Twins+SICL vs. Barlow Twins on DARai inertial). Please reconcile this number or clarify the calculation.
Circularity Check
No significant circularity: the central accuracy claims are evaluated on held-out subjects, and the only self-citations ([17], [18]) are non-load-bearing methodological references.
full rationale
The paper's central claim is that re-weighting same-subject negatives in the contrastive denominator improves generalization to unseen subjects. This claim is verified externally: Tables 1-4 report top-1 accuracy under the unseen-subject protocol on three public benchmarks (UTD-MHAD, MMAct, DARai), using frozen-encoder linear evaluation and cross-entropy fine-tuning (Sections 5.2-5.3). QSi is computed batch-wise from the model's own cosine similarities during training (Section 4.1), which is training signal rather than a quantity derived from test labels or test embeddings, so the reported gains are not forced by construction. The only self-references are [17] and [18] (same research group), cited for the negative-set decomposition ('Similar to [17, 18], we decompose our negatives into two sets') and for the batch-wise weighting scheme ('QSi is calculated batch-wise, similar to [17]'); these citations supply a construction template, not a proof of correctness, and the empirical claim does not depend on the cited papers' results being true. No uniqueness theorem is imported from the authors' prior work, and the method is not a renamed known result. The manuscript's genuine weakness is under-specification rather than circularity: QSi is described only in prose ('normalizing the exponentiated cosine similarities of same-subject negatives, summing these values, and then dividing by their average') with no closed-form equation, and the supervised analogue QSij in Eq. (5) is never defined at all; plausible alternative implementations would change which pairs are pushed apart, making the stated mechanism hard to attribute precisely. This is a reproducibility defect, not a reduction of the prediction to its inputs, because any implementation would still be scored on the same held-out subjects. Score 1 reflects the minor, non-load-bearing self-citation for the batch-wise weighting scheme, with no circular step identified.
Assumptions & free parameters
assumptions (3)
- domain assumption Subject identity is the dominant source of distribution shift in HAR sensor data.
- ad hoc to paper Amplifying the denominator contribution of same-subject negatives via QSi drives the encoder to discard subject-specific information while retaining activity information.
- domain assumption A 20-subject subset of DARai is representative enough for cross-subject evaluation.
Cite this review
Pith. "Pith review of Subject Invariant Contrastive Learning for Human Activity Recognition." pith.science (2026). https://pith.science/paper/6UM2Y6AO
@misc{pith2026250703250,
author = {Pith},
title = {Pith review of: Subject Invariant Contrastive Learning for Human Activity Recognition},
year = {2026},
howpublished = {\url{https://pith.science/paper/6UM2Y6AO}},
note = {Machine review of arXiv:2507.03250}
}
read the original abstract
The high cost of annotating data makes self-supervised approaches, such as contrastive learning methods, appealing for Human Activity Recognition (HAR). Effective contrastive learning relies on selecting informative positive and negative samples. However, HAR sensor signals are subject to significant domain shifts caused by subject variability. These domain shifts hinder model generalization to unseen subjects by embedding subject-specific variations rather than activity-specific features. As a result, human activity recognition models trained with contrastive learning often struggle to generalize to new subjects. We introduce Subject-Invariant Contrastive Learning (SICL), a simple yet effective loss function to improve generalization in human activity recognition. SICL re-weights negative pairs drawn from the same subject to suppress subject-specific cues and emphasize activity-specific information. We evaluate our loss function on three public benchmarks: UTD-MHAD, MMAct, and DARai. We show that SICL improves performance by up to 11% over traditional contrastive learning methods. Additionally, we demonstrate the adaptability of our loss function across various settings, including multiple self-supervised methods, multimodal scenarios, and supervised learning frameworks.
Forward citations
Cited by 1 Pith paper
-
Joint-Embedding Predictive Architecture for Sensor-based Activity Recognition
JEPA-based self-supervised pre-training on inertial sensor data improves recognition of rare transitional human activities compared to supervised learning, with gains mostly on transition classes.
Reference graph
Works this paper leans on
-
[17]
Vi- creg: Variance-invariance-covariance regularization for self- supervised learning,
Adrien Bardes, Jean Ponce, and Yann LeCun, “Vi- creg: Variance-invariance-covariance regularization for self- supervised learning,” arXiv preprint arXiv:2105.04906, 2021
arXiv 2021
-
[18]
Simcse: Simple contrastive learning of sentence embeddings,
Tianyu Gao, Xingcheng Yao, and Danqi Chen, “Simcse: Simple contrastive learning of sentence embeddings,” arXiv preprint arXiv:2104.08821, 2021
arXiv 2021
-
[1]
However, these applications rely on massive data drawn from various sensors
INTRODUCTION Human activity recognition (HAR) has broad applications in many areas, such as smart homes [1] and health monitoring [2]. However, these applications rely on massive data drawn from various sensors. Labeling this data presents a major challenge, as it is expensive and time consuming [3]. Furthermore, labeling time series data, such as acceler...
-
[2]
We introduce a simple and effective method to integrate sub- ject labels during contrastive training, making the representa- tions invariant to subject variations
-
[3]
We show that our approach outperforms traditional con- trastive methods and, when combined with those methods, further boosts their performance by making the learned rep- resentations subject-invariant
-
[4]
We show that our method can be used as a pre-training step for supervised training to increase generalization perfor- mance
-
[5]
We show that subject-invariant contrastive learning is also effective in supervised contrastive settings and recent mul- timodal HAR frameworks
-
[6]
RELA TED WORKS 2.1. Contrastive Learning and HAR Contrastive loss maps semantically similar (positive) samples closer in the embedding space while pushing dissimilar (negative) samples farther apart. This approach has proven effective at extracting mean- ingful representations from HAR data. For instance, one prominent approach [9] adapts the SimCLR frame...
Show all 35 references
-
[7]
For this analysis, we use the DARai [4] dataset, as it consists of daily activities collected in a natural, real- world setting from different subjects
SUBJECT V ARIABILITY IN HAR We first conduct an analysis to investigate whether the data acquired during human activities exhibits a domain shift problem due to subject-based variability. For this analysis, we use the DARai [4] dataset, as it consists of daily activities colle...
-
[8]
Our loss function selectively samples contrastive negatives based on their subject labels
METHODOLOGY We modify the traditional contrastive loss function to effectively uti- lize subject labels. Our loss function selectively samples contrastive negatives based on their subject labels. This encourages the model to learn distinctive characteristics of activities rath...
-
[9]
Datasets We evaluate the performance of our proposed approaches on three benchmark multimodal datasets: UTD-MHAD [22], MMAct [23], and DARai [4]
EXPERIMENTS AND RESULTS 5.1. Datasets We evaluate the performance of our proposed approaches on three benchmark multimodal datasets: UTD-MHAD [22], MMAct [23], and DARai [4]. UTD-MHAD [22] consists of recordings from 10 subjects per- forming 27 distinct activities, with each a...
-
[10]
CONCLUSION In this work, we introduce a contrastive loss function to leverage sub- ject labels to learn subject-invariant representations. Through exten- sive evaluation, we show our Subject-Invariant Contrastive Learn- ing (SICL) approach consistently improves generalization ...
-
[11]
Human activ- ity recognition in smart home with deep learning approach,
Homay Danaei Mehr and Huseyin Polat, “Human activ- ity recognition in smart home with deep learning approach,” in 2019 7th International Istanbul Smart Grids and Cities Congress and Fair (ICSG). IEEE, 2019, pp. 149–153
2019
-
[12]
Person- alizing human activity recognition models using incremental learning,
Pekka Siirtola, Heli Koskim ¨aki, and Juha R ¨oning, “Person- alizing human activity recognition models using incremental learning,” arXiv preprint arXiv:1905.12628, 2019
1905 arXiv
-
[13]
Focal: A cost- aware video dataset for active learning,
Kiran Kokilepersaud, Yash-Yee Logan, Ryan Benkert, Chen Zhou, Mohit Prabhushankar, Ghassan AlRegib, Enrique Corona, Kunjan Singh, and Mostafa Parchami, “Focal: A cost- aware video dataset for active learning,” in 2023 IEEE Inter- national Conference on Big Data (BigData) . IEE...
2023
-
[14]
Hierarchical and multimodal data for daily activity un- derstanding,
Ghazal Kaviani, Yavuz Yarici, Seulgi Kim, Mohit Prab- hushankar, Ghassan AlRegib, Mashhour Solh, and Ameya Patil, “Hierarchical and multimodal data for daily activity un- derstanding,” 2025
2025
-
[15]
A simple framework for contrastive learning of visual representations,
Ting Chen, Simon Kornblith, Mohammad Norouzi, and Ge- offrey Hinton, “A simple framework for contrastive learning of visual representations,” in International conference on ma- chine learning. PmLR, 2020, pp. 1597–1607
2020
-
[16]
Barlow twins: Self-supervised learning via redundancy reduction,
Jure Zbontar, Li Jing, Ishan Misra, Yann LeCun, and St ´ephane Deny, “Barlow twins: Self-supervised learning via redundancy reduction,” in International conference on machine learning . PMLR, 2021, pp. 12310–12320
2021
-
[19]
Contrastive self-supervised learning for sensor-based human activity recognition,
Bulat Khaertdinov, Esam Ghaleb, and Stylianos Asteriadis, “Contrastive self-supervised learning for sensor-based human activity recognition,” in 2021 IEEE International Joint Con- ference on Biometrics (IJCB). IEEE, 2021, pp. 1–8
2021
-
[20]
Multimodal contrastive learning with hard negative sampling for human ac- tivity recognition,
Hyeongju Choi, Apoorva Beedu, and Irfan Essa, “Multimodal contrastive learning with hard negative sampling for human ac- tivity recognition,” arXiv preprint arXiv:2309.01262, 2023
2023 arXiv
-
[21]
Contrastive learning with cross-modal knowledge mining for multimodal human activity recognition,
Razvan Brinzea, Bulat Khaertdinov, and Stylianos Asteriadis, “Contrastive learning with cross-modal knowledge mining for multimodal human activity recognition,” in2022 International Joint Conference on Neural Networks (IJCNN) . IEEE, 2022, pp. 01–08
2022
-
[22]
Cocoa: Cross modality contrastive learning for sensor data,
Shohreh Deldari, Hao Xue, Aaqib Saeed, Daniel V Smith, and Flora D Salim, “Cocoa: Cross modality contrastive learning for sensor data,” Proceedings of the ACM on Interactive, Mo- bile, Wearable and Ubiquitous Technologies, vol. 6, no. 3, pp. 1–28, 2022
2022
-
[23]
Subject vari- ability in sensor-based activity recognition,
Ali Olow Jimale and Mohd Halim Mohd Noor, “Subject vari- ability in sensor-based activity recognition,” Journal of Ambi- ent Intelligence and Humanized Computing, vol. 14, no. 4, pp. 3261–3274, 2023
2023
-
[24]
Generalizable sensor-based activity recogni- tion via categorical concept invariant learning,
Di Xiong, Shuoyuan Wang, Lei Zhang, Wenbo Huang, and Chaolei Han, “Generalizable sensor-based activity recogni- tion via categorical concept invariant learning,” arXiv preprint arXiv:2412.13594, 2024
2024 arXiv
-
[25]
Diversify: A gen- eral framework for time series out-of-distribution detection and generalization,
Wang Lu, Jindong Wang, Xinwei Sun, Yiqiang Chen, Xi- angyang Ji, Qiang Yang, and Xing Xie, “Diversify: A gen- eral framework for time series out-of-distribution detection and generalization,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 6, pp. 4...
2024
-
[26]
Latent independent excitation for generalizable sensor-based cross- person activity recognition,
Hangwei Qian, Sinno Jialin Pan, and Chunyan Miao, “Latent independent excitation for generalizable sensor-based cross- person activity recognition,” in Proceedings of the AAAI con- ference on artificial intelligence , 2021, vol. 35, pp. 11921– 11929
2021
-
[27]
Taxes are all you need: Integration of taxonomical hierarchy relationships into the contrastive loss,
Kiran Kokilepersaud, Yavuz Yarici, Mohit Prabhushankar, and Ghassan AlRegib, “Taxes are all you need: Integration of taxonomical hierarchy relationships into the contrastive loss,” arXiv preprint arXiv:2406.06848, 2024
2024 arXiv
-
[28]
Hex: Hierarchical emergence ex- ploitation in self-supervised algorithms,
Kiran Kokilepersaud, Seulgi Kim, Mohit Prabhushankar, and Ghassan AlRegib, “Hex: Hierarchical emergence ex- ploitation in self-supervised algorithms,” arXiv preprint arXiv:2410.23200, 2024
2024 arXiv
-
[29]
Supervised contrastive learning,
Prannay Khosla, Piotr Teterwak, Chen Wang, Aaron Sarna, Yonglong Tian, Phillip Isola, Aaron Maschinot, Ce Liu, and Dilip Krishnan, “Supervised contrastive learning,” Advances in neural information processing systems, vol. 33, pp. 18661– 18673, 2020
2020
-
[30]
Contrastive multiview coding,
Yonglong Tian, Dilip Krishnan, and Phillip Isola, “Contrastive multiview coding,” in Computer Vision–ECCV 2020: 16th Eu- ropean Conference, Glasgow, UK, August 23–28, 2020, Pro- ceedings, Part XI 16. Springer, 2020, pp. 776–794
2020
-
[31]
Contrastive learning with cross-modal knowledge mining for multimodal human activity recognition,
Razvan Brinzea, Bulat Khaertdinov, and Stylianos Asteriadis, “Contrastive learning with cross-modal knowledge mining for multimodal human activity recognition,” in2022 International Joint Conference on Neural Networks (IJCNN) . July 2022, p. 01–08, IEEE
2022
-
[32]
Utd- mhad: A multimodal dataset for human action recognition uti- lizing a depth camera and a wearable inertial sensor,
Chen Chen, Roozbeh Jafari, and Nasser Kehtarnavaz, “Utd- mhad: A multimodal dataset for human action recognition uti- lizing a depth camera and a wearable inertial sensor,” in 2015 IEEE International conference on image processing (ICIP) . IEEE, 2015, pp. 168–172
2015
-
[33]
Mmact: A large-scale dataset for cross modal human action understanding,
Quan Kong, Ziming Wu, Ziwei Deng, Martin Klinkigt, Bin Tong, and Tomokazu Murakami, “Mmact: A large-scale dataset for cross modal human action understanding,” in Pro- ceedings of the IEEE/CVF International Conference on Com- puter Vision (ICCV), October 2019
2019
-
[34]
Co- occurrence feature learning from skeleton data for action recognition and detection with hierarchical aggregation,
Chao Li, Qiaoyong Zhong, Di Xie, and Shiliang Pu, “Co- occurrence feature learning from skeleton data for action recognition and detection with hierarchical aggregation,”arXiv preprint arXiv:1804.06055, 2018
2018 arXiv
-
[35]
Contrastive learning with hard negative samples,
Joshua Robinson, Ching-Yao Chuang, Suvrit Sra, and Stefanie Jegelka, “Contrastive learning with hard negative samples,” arXiv preprint arXiv:2010.04592, 2020
2010 arXiv
Reviewed August 6, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.