Pith. sign in

REVIEW 4 major objections 5 minor 47 references

Transformer-Based Decomposition of Electrodermal Activity for Real-World Mental Health Applications

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper claims that a label-free Transformer, the Feel Transformer, decomposes wearable EDA into phasic and tonic components whose extracted SCR density, SCR amplitude, and tonic-slope direction agree with established methods, while…

desk verdict A plausible unsupervised EDA decomposition method, but the validity claim rests on a post hoc agreement analysis that needs a synthetic benchmark and a fixed protocol. read the letter →

arxiv 2506.06378 v1 pith:NSFFWVDG submitted 2025-06-04 eess.SP cs.LG

Charalampos Tsirmpas , Stasinos Konstantopoulos , Dimitris Andrikopoulos , Konstantina Kyriakouli , Panagiotis Fatouros This is my paper
classification eess.SPcs.LG
keywords electrodermalactivityphasicandtonicdecompositiontimeseriesdetrendingwearablesensorsTransformerself-supervisedlearningdigitalmentalhealthbiosignalprocessing
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Electrodermal activity (EDA), measured in microsiemens, is a window into the sympathetic nervous system, and researchers routinely split it into a fast phasic part (skin conductance responses, SCRs) and a slow tonic baseline (skin conductance level, SCL). The paper's central claim is that this split can be learned without supervision by "Feel Transformer," an Autoformer-based model whose tonic branch is constrained by pooling, and that the resulting decomposition is comparable to established methods on noisy, in-the-wild wearable data. Comparing on three consensus biomarkers—direction of SCL slope, density of SCR peaks, and mean SCR amplitude—the paper reports that Feel Transformer agrees with detrending on all three and with Ledalab, cvxEDA, and sparsEDA on peak density and amplitude. The authors treat this feature-level agreement as evidence of validity because no ground-truth decomposition exists, and they argue that Feel Transformer avoids the false phasic peaks that cubic interpolation creates when the signal has abrupt jumps.

What carries the argument

The load-bearing mechanism is the Feel Transformer's pooling-constrained additive decomposition, inherited from the Autoformer's trend-decomposition idea. The network runs one encoder and two decoder blocks non-autoregressively; an average-pooling layer produces the SCL component, and the residual between the pooled curve and the input becomes the phasic component, minimizing only reconstruction MSE. Because the pooling kernel sets the slowest timescale the tonic branch can follow, kernel sizes of 8x60+1 and 8x30+1 produce the reported agreement, while the 1-second kernel (8x1+1) lets the tonic curve hug the signal and leaves a noisy, low-amplitude phasic component. This mechanism encodes the domain knowledge that the tonic component is slow-moving without requiring labeled SCR examples.

What would settle it

Synthesize EDA signals with a known slow tonic baseline and sparse phasic impulses, then compare each method's recovered components against the ground truth; if the Feel Transformer's reconstruction error is large precisely on frames where it agrees with detrending rather than with the true components, the central claim would be refuted. A cheaper check is to fix the tonic-slope bin boundaries before running the comparison and see whether the reported agreement on SCL direction persists.

Watch

Extended reading notes

Core claim

The central discovery claimed is that a non-autoregressive Transformer, trained only to reconstruct 3-minute EDA windows with an average-pooling branch dedicated to the tonic component, learns a phasic/tonic separation whose extracted features match those of specialized EDA decompositions. The paper shows this empirically across configurations: with pooling kernels of 60 or 30 seconds, the model's tonic-slope histograms align with detrending, its SCR density aligns with all comparison methods, and its SCR amplitude aligns with sparsEDA and detrending at roughly 0.1 microsiemens, whereas cvxEDA's much higher mean amplitude is attributed to optimization artifacts. Visual inspection of characteristic frames shows that Ledalab and cvxEDA produce artificial peaks when SCL changes abruptly, while Feel Transformer compromises by tracking the abrupt change in the tonic component. The authors conclude that agreement on SCR features is a positive indication of validity, while SCL disagreements should be discounted because the comparison methods' SCL performance is itself unvalidated.

Load-bearing premise

The load-bearing premise is that agreement among imperfect decomposition methods on the three chosen features counts as evidence of validity; if all methods share the same biases, or if the post-hoc choice of tonic-slope bin boundaries manufactures the agreement, the comparative claim has no support.

Editorial extensions

If this is right

  • If the claimed agreement reflects true decomposition quality, wearable EDA can be processed in real time without per-user labels, manual peak fitting, or a deconvolution step.
  • On signals with abrupt SCL transitions, the method should produce fewer artifact peaks than cubic-interpolation methods, making it better suited to ambulatory and home monitoring.
  • Because the trained model is generative and non-autoregressive, it can be used to simulate or forecast future physiological states and then to extract additional features from the generated signal, beyond the three used for comparison.
  • The feature-agreement protocol itself—comparing SCL slope direction, SCR density, and SCR amplitude across methods—offers a practical evaluation route whenever no ground-truth decomposition exists.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A direct test of the validity assumption would be to generate synthetic EDA with a known slow baseline plus sparse impulses with steep onset and slow recovery, then measure how closely each method's recovered phasic and tonic components match the generating components; this would tell whether agreement among methods is a symptom of shared bias.
  • Because the tonic-slope bin boundaries at +/-0.001 microsiemens per second were chosen post hoc to make the pooled histogram uniform, re-running the comparison with boundaries fixed a priori (or with continuous metrics) would show whether the reported agreement on SCL direction is robust.
  • The same pooling-constrained autoencoder idea could be applied to other ambulatory biosignals that have a slow trend plus fast events, such as heart-rate variability or actigraphy, with the same no-ground-truth caveat.
  • Since the evaluation uses one 40-person longitudinal dataset from one device ecosystem, a transfer test across sensor generations and sampling rates would establish whether the learned decomposition generalizes without retraining.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. This paper presents the Feel Transformer, a Transformer-based model adapted from the Autoformer architecture, for decomposing electrodermal activity (EDA) into phasic and tonic components without explicit supervision. The model uses average pooling and a reconstruction loss to bias one output toward a slow-moving tonic component and the other toward a fast-moving phasic component. The authors evaluate three pooling-kernel configurations on in-the-wild data collected from 40 individuals over 16 weeks, comparing the extracted features (tonic slope direction, SCR density, SCR amplitude) with those produced by Ledalab, cvxEDA, sparsEDA, and Theil detrending. Because no ground-truth decomposition exists, the paper uses agreement between methods as a proxy for validity, and argues that agreement with domain-specific methods on the SCR features is a positive indication of the Feel Transformer's validity. The paper also presents qualitative examples suggesting that Ledalab and cvxEDA produce spurious peaks due to cubic interpolation artifacts in signals with abrupt changes.

Significance. If the central empirical claim were fully supported, the paper would offer a practically useful contribution: an unsupervised, domain-agnostic deep-learning decomposition method that is robust to noisy wearable data and suitable for real-time mental-health applications. The authors are transparent about the absence of ground truth, and the study uses a comparatively large longitudinal in-the-wild dataset. The core training objective is unsupervised reconstruction rather than fitting to the reported features, so there is no direct derivation-to-fit circularity. However, the validity evidence is substantially weakened by post hoc analysis choices, by selective treatment of agreement versus disagreement, and by the absence of statistical uncertainty quantification. These issues currently make the headline comparative claim difficult to evaluate as written.

major comments (4)
  1. [Section 5.1, Table 1] The tonic-slope bin boundaries at ±0.001 µS/s are selected 'because they give a uniform histogram when bundling together the tonic components from all decomposition methods.' This is a post hoc calibration on the same data used to report agreement, so each method's distance from a 33/33/33 split is not an independent measurement. The claim that Theil and Feel Transformer give results 'similar to each other' is therefore not a tested conclusion. Please fix the boundaries a priori, report sensitivity to the chosen boundaries, and clarify whether the uniform pooled histogram is intended as a sanity check or as a benchmark.
  2. [Section 6 vs. Section 3] The paper first identifies SCL direction, SCR density, and SCR amplitude as the three consensus features for comparing decompositions (Section 3), but then dismisses the SCL-direction disagreement between Feel Transformer and Ledalab/cvxEDA/sparsEDA with the argument that domain-specific methods' SCL performance 'has not been validated' (Section 6). If agreement on a feature is treated as positive validity evidence, disagreement on the same feature must carry comparable evidential weight; the asymmetry is unjustified. The comparison would be more credible with a pre-specified set of features and a symmetric decision rule, or with all three features reported without post hoc exclusions.
  3. [Section 5.1, Tables 1-3] All comparisons are presented as point estimates without confidence intervals, standard errors, or statistical tests. The data come from 40 subjects over 16 weeks, and the non-overlapping 3-min frames within a subject are not independent, so apparent percentage differences across methods could easily arise by chance. Please report subject-level or clustered uncertainty (for example, bootstrap over subjects) and, where relevant, tests of distributional differences. This is necessary to support statements such as 'Ledalab, cxvEDA, detrending, and the first two configurations of the Feel Transformer generally agree on peak density' and 'cvxEDA stands out in giving considerably higher amplitudes.'
  4. [Section 4.2] The training protocol for the Feel Transformer is underspecified. The manuscript does not report the number of epochs, learning rate, optimizer, batch size, subject-wise train/validation/test split, or the procedure by which the three pooling-kernel configurations were compared. The first configuration is justified only indirectly, and the third configuration is discarded after noting that its output 'is not desired' (Section 5.1), which is a post hoc model-selection criterion. Without a complete protocol or a code release, the central methodological contribution cannot be reproduced or assessed for overfitting.
minor comments (5)
  1. [Section 1] The sentence 'Given the distinct physiological information each component provides, it becomes evident that decomposing EDA signals into their phasic and tonic components is critical for a comprehensive understanding of ANS functioning.' is repeated verbatim in the introduction.
  2. [Throughout] There are several typos and inconsistent spellings: 'cxvEDA' in Table 2, 'cvxLeda' in Section 6, 'Leadalab' in Section 5.1, and 'Tsirbas' in reference [43] where the authors' name is otherwise given as 'Tsirmpas.' These should be corrected.
  3. [Footnote 1] The corresponding-author email in the footnote contains garbled characters ('envel⌢pe-⌢penharis'), which should be fixed to a valid email address.
  4. [Section 4.1] The manuscript does not include a data or code availability statement. Given that the dataset is proprietary, at least the training procedure, configuration files, and feature-extraction scripts should be released to make the comparison reproducible.
  5. [Section 4.2] The relationship between the embedding dimension, attention heads, and pooling kernel sizes is described in prose; a small table or pseudocode block would make the architecture easier to follow and reproduce.

Circularity Check

2 steps flagged · score 3.0 of 10

Post hoc bin calibration and selectively applied agreement evidence make the validity claim partly self-supporting, although the core unsupervised reconstruction is not circular.

  1. fitted input called prediction [Section 5.1, Table 1 (tonic slope binning)]
    "The bin boundaries are set at −0.001𝜇𝑆/𝑠𝑒𝑐 and 0.001𝜇𝑆/𝑠𝑒𝑐. These boundaries were selected because they give a uniform histogram when bundling together the tonic components from all decomposition methods. We then calculate the histograms separately for the tonic components produced by each method. By comparing these histograms against a uniform 33% – 33% – 33% split, we can see any possible bias towards exaggerating or understating the tonic slope."

    The reported falling/stable/rising percentages depend on bin edges that were calibrated on the pooled tonic slopes of all methods to make the pooled histogram uniform. Therefore the 33/33/33 baseline is not an independent reference: the average histogram is uniform by construction, and the per-method percentages are summarized against a yardstick fitted to the very same pooled data. The SCL-direction agreement is thus a post hoc calibration result, not an independent test of concordance among methods.

  2. other [Section 6, Conclusions]
    "we observe that the agreement of Feel Transformer with domain-specific methods on the SCR features is a positive indication for the validity of our method; While the minor disagreement between Feel Transformer and domain-specific methods on the SCL should not be taken into account since the performance of domain-specific methods on the SCL features has not been validated."

    The validity argument admits agreement with domain methods as positive evidence but declares the only feature where Feel Transformer disagrees with those methods inadmissible. Since the domain methods are equally unvalidated on SCR features, there is no ex ante rule that distinguishes the admissible agreement from the inadmissible disagreement. The conclusion is protected from disconfirmation by selectively defining which agreement counts, making the central comparative claim unfalsifiable as written.

full rationale

The core Feel Transformer derivation is not circular: the model is trained with an unsupervised MSE reconstruction loss, and the phasic/tonic features used for comparison are computed after training, not used as supervision. The data are drawn from the authors' earlier studies, but that is a normal data-source citation and is not load-bearing in the derivation. The Autoformer base and the cited EDA methods are independent prior work. The two issues that raise the score are the post hoc selection of the tonic-slope bin boundaries on the pooled data (Table 1) and the selective treatment of agreement versus disagreement in Section 6. These do not make the reconstruction itself equivalent to its inputs, but they make the reported agreement on SCL direction partly manufactured and render the overall validity claim unfalsifiable. The choice to downweight the third pooling configuration is likewise made after inspecting results, but since all configurations are reported, it is a transparent post hoc model selection rather than a circular derivation. The score is therefore moderate, not severe.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The paper's comparative claim rests on several hand-chosen settings: the SCL pooling kernel size (three variants, with one selected as preferred), the tonic slope bin boundaries (calibrated to the pooled data), and the transformer hyperparameters. It also relies on domain assumptions about SCR sparsity, the Bateman impulse shape, and the sufficiency of three features as a proxy for decomposition quality. No external ground truth or independent validation is used.

free parameters (4)
  • SCL average pooling kernel size = 8x60+1 (FT1), 8x30+1 (FT2), 8x1+1 (FT3); FT2 preferred
    Authors test three kernel sizes and select FT2 based on qualitative expectations; the third is dismissed as 'not desired' (Section 5.1), so the choice is not grounded in an external criterion.
  • Tonic slope bin boundaries = +/- 0.001 uS/sec
    Boundaries selected because they give a uniform histogram when bundling together tonic components from all decomposition methods (Section 5.1); this is a post hoc choice that shapes the reported comparisons.
  • Transformer hyperparameters = embedding dim 32, 4 heads, 1 encoder, 2 decoder layers, first conv output 16
    Hyperparameters are stated without a search or ablation against an external benchmark; they are chosen by the authors.
  • Low-pass filter cutoff = 3 Hz, 4th-order Butterworth
    Standard preprocessing choice applied to all methods; affects all results but is not varied or justified against alternatives.
assumptions (4)
  • domain assumption SCRs are sparse and do not drastically affect the estimation of the tonic component.
    Invoked to justify comparing Feel Transformer with detrending and to justify the pooling-based tonic estimate; see Section 6: 'under the (reasonable) assumption that SCRs are sparse...'.
  • domain assumption The Bateman biexponential function captures the physiology of phasic impulses.
    Borrowed from previous work (Benedek and Kaernbach) and used to characterize Ledalab and cvxEDA; the paper does not re-derive it.
  • ad hoc to paper The three features (SCL direction, SCR frequency, SCR amplitude) are the appropriate similarity metrics for decomposition quality.
    Chosen because they are the only features extractable from all methods (Section 3); no external validation that these features are sufficient for clinical or practical utility.
  • ad hoc to paper MSE reconstruction loss plus pooling bias yields physiologically meaningful phasic/tonic separation.
    The paper argues the pooling forces the SCL to be slow-moving, but provides no proof or external benchmark that the resulting decomposition corresponds to physiology (Sections 4.2, 5, 6).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Transformer-Based Decomposition of Electrodermal Activity for Real-World Mental Health Applications." pith.science (2026). https://pith.science/paper/NSFFWVDG

@misc{pith2026250606378,
  author       = {Pith},
  title        = {Pith review of: Transformer-Based Decomposition of Electrodermal Activity for Real-World Mental Health Applications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NSFFWVDG}},
  note         = {Machine review of arXiv:2506.06378}
}
read the original abstract

Decomposing Electrodermal Activity (EDA) into phasic (short-term, stimulus-linked responses) and tonic (longer-term baseline) components is essential for extracting meaningful emotional and physiological biomarkers. This study presents a comparative analysis of knowledge-driven, statistical, and deep learning-based methods for EDA signal decomposition, with a focus on in-the-wild data collected from wearable devices. In particular, the authors introduce the Feel Transformer, a novel Transformer-based model adapted from the Autoformer architecture, designed to separate phasic and tonic components without explicit supervision. The model leverages pooling and trend-removal mechanisms to enforce physiologically meaningful decompositions. Comparative experiments against methods such as Ledalab, cvxEDA, and conventional detrending show that the Feel Transformer achieves a balance between feature fidelity (SCR frequency, amplitude, and tonic slope) and robustness to noisy, real-world data. The model demonstrates potential for real-time biosignal analysis and future applications in stress prediction, digital mental health interventions, and physiological forecasting.

Figures

Figures reproduced from arXiv: 2506.06378 by the authors.

Figure 1
Figure 1. A characteristic example of false peaks created by the forced smoothness of cubic interpolation even when the original signal has abrupt changes. EDA signal (black) tonic component (green) are shown on the left-hand side; the phasic component and the peaks detected are shown on the right-hand side. Since the phasic component is the residue between the EDA signal and the tonic component, its range (max-min amplitude)… view at source ↗
Figure 2
Figure 2. A characteristic example of false peaks created by cvxEDA. EDA signal (black) tonic component (green) are shown on the left-hand side; the phasic component and the peaks detected are shown on the right-hand side. reach the lower SCL level after 215sec, and this steep SCL slope creates the rise in phasic at 140-150sec that creates the peak. When using a finer kernel to allow the SCL to more closely follow the EDA, th… view at source ↗
Figure 3
Figure 3. Comparison of the three configurations of the Feel Transformer. EDA signal (black) tonic component (green) are shown on the left-hand side; the phasic component and the peaks detected are shown on the right￾hand side. are usually acquired, but are a lot more common in personal healthcare and well-being applications that rely on the EDA signal acquired in the wild from wearable devices. It should also be noted that l… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

47 extracted references · 33 canonical work pages

  1. [1]

    Sensing Technologies for Monitoring Serious Mental Illnesses.IEEE MultiMedia2018,25, 61–75

    Abdullah, S.; Choudhury, T. Sensing Technologies for Monitoring Serious Mental Illnesses.IEEE MultiMedia2018,25, 61–75. https://doi.org/10.1109/MMUL.2018.011921236

    work page arXiv 2018

  2. [2]

    Sano, A.; Taylor, S.; McHill, A.W.; Phillips, A.J.; Barger, L.K.; Klerman, E.; Picard, R. Identifying Objective Physiological Markers and Modifiable Behaviors for Self-Reported Stress and Mental Health Status Using Wearable Sensors and Mobile Phones: Observational Study.J Med Internet Res2018,20, e210. https://doi.org/10.2196/jmir.9410

    doi:10.2196/jmir.9410

  3. [3]

    Singh, J.; Sharma, D. Automated detection of mental disorders using physiological signals and machine learning: A systematic review and scientometric analysis.Multimedia Tools and Applications2024,83, 73329–73361. https://doi.org/10.1007/s11042-023-17504-1

    work page doi:10.1007/s11042-023-17504-1

  4. [4]

    https://doi.org/10

    Boucsein, W.Electrodermal Activity; Springer Science & Business Media, 2012. https://doi.org/10. 1007/978-1-4614-1126-0

    work page 2012

  5. [5]

    The Electrodermal System

    Dawson, M.E.; Schell, A.M.; Filion, D.L. The Electrodermal System. InIn Handbook of Psychophysi- ology (3rd ed.); Cambridge University Press, 2012. https://doi.org/10.1017/CBO9780511546396.007

    work page doi:10.1017/cbo9780511546396.007 2012

  6. [6]

    Is Mediation of Sweating Cholinergic, Adrenergic, or Both? A Comment on the Literature.Psychophysiology1987,24

    Shields, S.A.; MacDowell, K.A.; Fairchild, S.B.; Campbell, M.L. Is Mediation of Sweating Cholinergic, Adrenergic, or Both? A Comment on the Literature.Psychophysiology1987,24. https://doi.org/10.1111/j.1469-8986.1987.tb00301.x

    arXiv 1987

  7. [7]

    Electro- dermal Activity for Measuring Cognitive and Emotional Stress Level.Journal of Medical Signals & Sensors2022,12, 155–162

    Rahma, O.N.; Putra, A.P.; Rahmatillah, A.; Putri, Y.S.K.A.; Fajriaty, N.D.; Ain, K.; Chai, R. Electro- dermal Activity for Measuring Cognitive and Emotional Stress Level.Journal of Medical Signals & Sensors2022,12, 155–162. https://doi.org/10.4103/jmss.JMSS_78_20

    work page doi:10.4103/jmss.jmss_78_20

  8. [8]

    Elec- trodermal Activity as a Measure of Cognitive Load: A Methodological Approach

    Buchwald, M.; Kupiński, S.; Bykowski, A.; Marcinkowska, J.; Ratajczyk, D.; Jukiewicz, M. Elec- trodermal Activity as a Measure of Cognitive Load: A Methodological Approach. In Proceed- ings of the 23rd International Conference on Signal Processing: Algorithms, Architectures, Arrangements, and Applications (SPA 2019), Poznan, Poland, 18-20 Sep 2019, 2019, ...

    work page arXiv 2019

Show all 47 references
  1. [9]

    https://doi.org/10.1007/978-3-319-46705-4

    Greco, A.; Valenza, G.; Scilingo, E.P.Advances in Electrodermal Activity Processing with Applications for Mental Health; Springer Nature, 2016. https://doi.org/10.1007/978-3-319-46705-4

    2016 doi

  2. [10]

    Monitoring Device, 2024

    Therapeutics, F. Monitoring Device, 2024. Accessed: 2024-05-15

    2024

  3. [11]

    E4: Wearable Physiological Sensing Platform, 2024

    Empatica. E4: Wearable Physiological Sensing Platform, 2024. Accessed: 2024-05-15

    2024

  4. [12]

    Shimmer3 GSR Unit, 2024

    Sensing, S. Shimmer3 GSR Unit, 2024. Accessed: 2024-05-15

    2024

  5. [13]

    Milstein, N.; Gordon, I. Validating Measures of Electrodermal Activity and Heart Rate Variability Derived From the Empatica E4 Utilized in Research Settings That Involve Interactive Dyadic States.Frontiers in Behavioral Neuroscience2020,Volume 14 - 2020. https://doi.org/10.338...

    2020

  6. [14]

    Wearable Emotion Recognition System based on GSR and PPG Signals

    Udovičić, G.; Ðerek, J.; Russo, M.; Sikora, M. Wearable Emotion Recognition System based on GSR and PPG Signals. In Proceedings of the Proceedings of the 2nd International Workshop on Multimedia for Personal Health and Health Care, New York, NY, USA, 2017; MMHealth ’17, p. 53–...

    2017

  7. [15]

    Randomized controlled study of a digital data driven intervention for depressive and generalized anxiety symptoms.npj Digital Medicine2025, 8, 113

    Fatouros, P.; Tsirmpas, C.; Andrikopoulos, D.; et al. Randomized controlled study of a digital data driven intervention for depressive and generalized anxiety symptoms.npj Digital Medicine2025, 8, 113. https://doi.org/10.1038/s41746-025-01511-7

    doi

  8. [16]

    Machine learning-enabled detection of attention-deficit/hyperactivity disorder with multimodal physiological data: a case-control study

    Andrikopoulos, D.; Vassiliou, G.; Fatouros, P.; et al. Machine learning-enabled detection of attention-deficit/hyperactivity disorder with multimodal physiological data: a case-control study. BMC Psychiatry2024,24, 547. https://doi.org/10.1186/s12888-024-05987-7

    doi

  9. [18]

    Assessment of Resilience of the Hellenic Navy Seals by Electrodermal Activity during Cognitive Tasks.International Journal of Environmental Research and Public Health2021,18

    Mourtakos, S.; Vassiliou, G.; Kontoangelos, K.; Papageorgiou, C.; Philippou, A.; Bersimis, F.; Geladas, N.; Koutsilieris, M.; Sidossis, L.S.; Tsirmpas, C.; et al. Assessment of Resilience of the Hellenic Navy Seals by Electrodermal Activity during Cognitive Tasks.International...

    doi

  10. [19]

    Does Wearable-Measured Heart Rate Variability During Sleep Predict Perceived Morning Mental and Physical Fitness?Applied Psychophysiology and Biofeedback2023,48, 247–257

    de Vries, H.; Oldenhuis, H.; van der Schans, C.; et al. Does Wearable-Measured Heart Rate Variability During Sleep Predict Perceived Morning Mental and Physical Fitness?Applied Psychophysiology and Biofeedback2023,48, 247–257. https://doi.org/10.1007/s10484-022-09578-8

    doi

  11. [20]

    Physical and mental health in adolescence: novel insights from a transdiagnostic examination of FitBit data in the ABCD study.Translational Psychiatry 2024,14, 75

    Damme, K.; Vargas, T.; Walther, S.; et al. Physical and mental health in adolescence: novel insights from a transdiagnostic examination of FitBit data in the ABCD study.Translational Psychiatry 2024,14, 75. https://doi.org/10.1038/s41398-024-02794-2

    2024 doi

  12. [21]

    A wearable sensor for unobtrusive, long-term assessment of electrodermal activity.IEEE Transactions on Biomedical Engineering2010,57, 1243–1252

    Poh, M.Z.; Swenson, N.C.; Picard, R.W. A wearable sensor for unobtrusive, long-term assessment of electrodermal activity.IEEE Transactions on Biomedical Engineering2010,57, 1243–1252. https://doi.org/10.1109/TBME.2009.2038487

    2009

  13. [22]

    Model-based analysis of skin conductance responses: Towards causal models in psychophysiology.Psychophysiology2013,50

    Bach, D.R.; Friston, K.J. Model-based analysis of skin conductance responses: Towards causal models in psychophysiology.Psychophysiology2013,50. https://doi.org/10.1111/j.1469-8986.2012. 01483.x

    2012

  14. [23]

    Decomposition of Skin Conductance Data by Means of Nonnegative Deconvolution.Psychophysiology2010,47

    Benedek, M.; Kaernbach, C. Decomposition of Skin Conductance Data by Means of Nonnegative Deconvolution.Psychophysiology2010,47. https://doi.org/10.1111/j.1469-8986.2009.00972.x

    2009

  15. [24]

    A Continuous Measure of Phasic Electrodermal Activity.Journal of Neuroscience Methods2010,190

    Benedek, M.; Kaernbach, C. A Continuous Measure of Phasic Electrodermal Activity.Journal of Neuroscience Methods2010,190. https://doi.org/10.1016/j.jneumeth.2010.04.028

    2010 doi

  16. [25]

    Separating individual skin conductance re- sponses in a short interstimulus-interval paradigm.Journal of Neuroscience Methods2005,146

    Alexander, D.M.; Trengove, C.; Johnston, P.; Cooper, T.; August, J.P.; Gordon, E. Separating individual skin conductance re- sponses in a short interstimulus-interval paradigm.Journal of Neuroscience Methods2005,146. https://doi.org/10.1016/j.jneumeth.2005.02.001

    doi

  17. [26]

    cvxEDA: A Convex Optimization Approach to Electrodermal Activity Processing.IEEE Trans Biomed Eng2016,63

    Greco, A.; Valenza, G.; Lanata, A.; Scilingo, E.P.; Citi, L. cvxEDA: A Convex Optimization Approach to Electrodermal Activity Processing.IEEE Trans Biomed Eng2016,63. https://doi.org/ 10.1109/TBME.2015.2474131

    2015

  18. [27]

    Feature Extraction of Galvanic Skin Re- sponses by Nonnegative Sparse Deconvolution.IEEE Journal of Biomedical and Health Informatics 2018,22

    Hernando-Gallego, F.; Luengo, D.; Artés-Rodríguez, A. Feature Extraction of Galvanic Skin Re- sponses by Nonnegative Sparse Deconvolution.IEEE Journal of Biomedical and Health Informatics 2018,22. https://doi.org/10.1109/JBHI.2017.2780252

    2018

  19. [28]

    Skin Conductance as a Viable Alternative for Closing the Deep Brain Stimulation Loop in Neuropsychiatric Disorders.Frontiers in Neuroscience 2019,13

    Wickramasuriya, D.S.; Amin, M.R.; Faghih, R.T. Skin Conductance as a Viable Alternative for Closing the Deep Brain Stimulation Loop in Neuropsychiatric Disorders.Frontiers in Neuroscience 2019,13. https://doi.org/10.3389/fnins.2019.00780

    2019

  20. [29]

    Improved M-FOCUSS Algorithm With Overlapping Blocks for Locally Smooth Sparse Signals.IEEE Transactions on Signal Processing2008,56

    Zdunek, R.; Cichocki, A. Improved M-FOCUSS Algorithm With Overlapping Blocks for Locally Smooth Sparse Signals.IEEE Transactions on Signal Processing2008,56. https://doi.org/10.1109/ TSP.2008.928160

    2008

  21. [30]

    A Compressed Sensing Based Decomposition of Electrodermal Activity Signals.IEEE Trans Biomed Eng2017,64

    Jain, S.; Oswal, U.; Xu, K.S.; Eriksson, B.; Haupt, J. A Compressed Sensing Based Decomposition of Electrodermal Activity Signals.IEEE Trans Biomed Eng2017,64. https://doi.org/10.1109/TBME. 2016.2632523

    2016

  22. [31]

    Goodfellow, I.; Bengio, Y.; Courville, A.Deep Learning; MIT Press, 2016

    2016

  23. [32]

    Long short-term memory.Neural computation1997,9, 1735–1780

    Hochreiter, S.; Schmidhuber, J. Long short-term memory.Neural computation1997,9, 1735–1780

  24. [33]

    Recurrent Neural Networks for Time Series Forecast- ing: Current status and future directions.International Journal of Forecasting2021,37, 388–427

    Hewamalage, H.; Bergmeir, C.; Bandara, K. Recurrent Neural Networks for Time Series Forecast- ing: Current status and future directions.International Journal of Forecasting2021,37, 388–427. https://doi.org/10.1016/j.ijforecast.2020.06.008

    doi

  25. [34]

    Conditional time series forecasting with convolutional neural networks.arXiv preprint arXiv:1703.046912017

    Borovykh, A.; Bohte, S.; Oosterlee, C.W. Conditional time series forecasting with convolutional neural networks.arXiv preprint arXiv:1703.046912017. https://doi.org/10.48550/arXiv.1703.04691

    doi arXiv

  26. [35]

    Multiple convolutional neural networks for multivariate time series prediction.Neurocomputing2019,367, 189–197

    Wang, K.; Li, K.; Zhou, L.; Hu, Y.; Cheng, Z.; Liu, J.; Chen, C. Multiple convolutional neural networks for multivariate time series prediction.Neurocomputing2019,367, 189–197. https: //doi.org/10.1016/j.neucom.2019.05.023

    2019 doi

  27. [36]

    Attention is all you need

    Vaswani, A.; Shazeer, N.; Parmar, N.; Uszkoreit, J.; Jones, L.; Gomez, A.N.; Kaiser, Ł.; Polosukhin, I. Attention is all you need. In Proceedings of the Advances in neural information processing systems, 2017, pp. 5998–6008

    2017

  28. [37]

    BERT: Pre-training of deep bidirectional transform- ers for language understanding

    Devlin, J.; Chang, M.W.; Lee, K.; Toutanova, K. BERT: Pre-training of deep bidirectional transform- ers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies...

    2019

  29. [38]

    Informer: Beyond effi- cient transformer for long sequence time-series forecasting

    Zhou, H.; Zhang, S.; Peng, J.; Zhang, S.; Li, J.; Xiong, H.; Zhang, W. Informer: Beyond effi- cient transformer for long sequence time-series forecasting. In Proceedings of the 35th AAAI Conference on Artificial Intelligence. AAAI Press, 2021

    2021

  30. [39]

    Reformer: The Efficient Transformer

    Kitaev, N.; Kaiser, Ł.; Levskaya, A. Reformer: The Efficient Transformer. In Proceedings of the International Conference on Learning Representations (ICLR), 2020

    2020

  31. [40]

    Autoformer: Decomposition Transformers with Auto-correlation for Long-Term Series Forecasting

    Wu, H.; Xiao, J.; Wang, W.; Long, M.; Wang, J. Autoformer: Decomposition Transformers with Auto-correlation for Long-Term Series Forecasting. In Proceedings of the Advances in Neural Information Processing Systems 34 (NeurIPS 2021), 2021

    2021

  32. [41]

    Autonomic Nervous System Activity in Emotion: A Review.Biological Psychology 2010,84

    Kreibig, S.D. Autonomic Nervous System Activity in Emotion: A Review.Biological Psychology 2010,84. https://doi.org/10.1016/j.biopsycho.2010.03.010

    2010 doi

  33. [42]

    Feature Extraction for Psychophysiologi- cal Load Assessment in Unconstrained Scenarios

    Silva, H.; Fred, A.; Eusé’bio, S.; Torrado, M.; Ouakinin, S. Feature Extraction for Psychophysiologi- cal Load Assessment in Unconstrained Scenarios. In Proceedings of the 2012 Annual International Conference of the IEEE Engineering in Medicine and Biology Society, San Diego, ...

    2012

  34. [43]

    Feasibility, engagement, and preliminary clinical outcomes of a digital biodata- driven intervention for anxiety and depression.Frontiers in Digital Health2022,4

    Tsirbas, C.; Andrikopoulos, D.; Fatouros, P.; Eleftheriou, G.; Anguera, J.A.; Kontoangelos, K.; Papageorgiou, C. Feasibility, engagement, and preliminary clinical outcomes of a digital biodata- driven intervention for anxiety and depression.Frontiers in Digital Health2022,4. h...

  35. [44]

    A Digital Mental Health Support Program for Depression and Anxiety in Populations With Attention-Deficit/Hyperactivity Disorder: Feasibility and Usability Study.JMIR Form Res.2023

    Tsirmpas, C.; Nikolakopoulou, M.; Kaplow, S.; Andrikopoulos, D.; Fatouros, P.; Kontoangelos, K.; Papageorgiou, C. A Digital Mental Health Support Program for Depression and Anxiety in Populations With Attention-Deficit/Hyperactivity Disorder: Feasibility and Usability Study.JM...

    2023 doi

  36. [45]

    Autoformer: Decomposition Transformers with Auto- Correlation for Long-Term Series Forecasting

    Wu, H.; Xu, J.; Wang, J.; Long, M. Autoformer: Decomposition Transformers with Auto- Correlation for Long-Term Series Forecasting. In Proceedings of the Advances in Neural Infor- mation Processing Systems 34 (NeurIPS 2021), 2021

    2021

  37. [46]

    Integrating wearable sensor data and self-reported diaries for personalized affect forecasting

    Yang, Z.; Wang, Y.; Yamashita, K.S.; Khatibi, E.; Azimi, I.; Dutt, N.; Borelli, J.L.; Rahmani, A.M. Integrating wearable sensor data and self-reported diaries for personalized affect forecasting. Smart Health2024,32, 100464. https://doi.org/https://doi.org/10.1016/j.smhl.2024.100464

  38. [47]

    The Role of Machine Learning in AR/VR-Based Cognitive Therapies: A Systematic Review for Mental Health Disorders.Electronics2025,14

    Halkiopoulos, C.; Gkintoni, E. The Role of Machine Learning in AR/VR-Based Cognitive Therapies: A Systematic Review for Mental Health Disorders.Electronics2025,14. https://doi.org/10.3390/ electronics14061110

  39. [2012]

    https://doi.org/10.1109/EMBC.2012.6347037

    2012

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.