Pith. sign in

REVIEW 4 major objections 6 minor 51 references

Non-invasive Blood Glucose Estimation from Wearable Physiological Signals

T0 review · 4 major / 6 minor · reviewed 2026-07-11 · grok-4.5

Pith's one-line read A continual-learning pipeline estimates blood glucose from PPG with 0.64 mmol/L MAE on a 183-person longitudinal wearable benchmark.

desk verdict Solid methods-and-benchmark paper: real longitudinal PPG–BG corpus plus a carefully isolated continual-learning pipeline; headline window metrics are a bit soft, but aggregation and ablations keep the claim honest enough for referees. read the letter →

arxiv 2607.04414 v1 pith:QBM6QGSX submitted 2026-07-05 eess.SP

classification eess.SP
keywords non-invasivebloodglucosephotoplethysmographydynamicincrementallearningcontinualproxygradientsmutualentropyclusteringwearablephysiologicalsignalslongitudinalbenchmark
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

Fingerstick and CGM glucose checks are invasive and costly, while wearable PPG is easy to collect but drifts over time and usually lacks matching glucose labels. This paper argues that the right learning setup can still turn short PPG windows into usable glucose estimates: discover latent physiological regimes, replay them under gradient constraints so older knowledge is not erased, and train a proxy-gradient agent so unlabeled PPG streams help the model adapt without inventing fake glucose labels. The authors also release a 183-participant, 285-day fingertip PPG, capillary glucose, and cuff blood-pressure benchmark for the community. On subject-independent five-fold validation the full system reaches 0.64 mmol/L mean absolute error, 1.29 mmol/L RMSE, and about 98% of estimates in Clarke zones A+B, remaining stable when windows are averaged by acquisition or by subject. The practical claim is a proof-of-concept that adaptive, label-efficient wearable glucose estimation is possible under real longitudinal drift, not only on static snapshots.

What carries the argument

Dynamic Incremental Learning (DIL) with MERDC and PGBA: ME2AC partitions streaming PPG into latent tasks by balancing local signal entropy and inter-cluster mutual information for task-balanced replay and gradient projection, while PGBA maps model outputs to bounded proxy gradients so unlabeled PPG windows update the model without pseudo-glucose labels.

What would settle it

On held-out people or multi-center wearable streams, if the full DIL+MERDC+PGBA system fails to stay near 0.64 mmol/L MAE with high Clarke A+B under subject-independent evaluation—especially if ablating MERDC or PGBA no longer hurts, or high-glucose and comorbidity subgroups collapse—the central claim fails.

Watch

Extended reading notes

Core claim

Dynamic incremental learning that pairs mutual-entropy task discovery and replay (MERDC/ME2AC) with an uncertainty-quantified Proxy Gradient Bridging Agent can continually adapt a PPG-to-glucose regressor under distribution drift and sparse labels, reaching 0.64 ± 0.01 mmol/L MAE and 97.69 ± 1.63% Clarke A+B on a new 183-participant longitudinal benchmark under subject-independent validation.

Load-bearing premise

The claim stands only if short PPG waveform windows carry a stable, learnable link to capillary glucose that can be split into a few unsupervised latent tasks whose replay and bounded proxy gradients are enough to keep adapting without erasing older knowledge.

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 / 6 minor

Summary. The manuscript proposes Dynamic Incremental Learning (DIL) for non-invasive blood glucose estimation from PPG, combining MERDC/ME2AC for unsupervised latent-task discovery and replay with a Proxy Gradient Bridging Agent (PGBA) that exploits unlabeled PPG via bounded proxy gradients. It also introduces a longitudinal benchmark of 183 participants (5,224 four-second windows) with synchronized PPG, capillary BG, and cuff BP over 285 days. Under subject-independent 5-fold CV with explicit fold isolation, the full system reports MAE 0.64±0.01 mmol/L, RMSE 1.29±0.10 mmol/L, and 97.69±1.63% in Clarke A+B, outperforming reimplemented baselines under a fixed-configuration protocol. Ablations, a PGBA alignment audit, aggregation-level ISO-style analyses, glucose-range/temporal slices, and a small 7-day pilot support a proof-of-concept claim for label-efficient adaptation under longitudinal drift.

Significance. If the results hold under the stated protocol, the work is a useful contribution on two fronts: (i) a carefully curated longitudinal PPG–capillary-BG benchmark with transparent curation and leakage controls, which the field currently lacks at this scale; and (ii) a domain-adapted continual-regression recipe that jointly addresses distribution drift and label scarcity, with component ablations and a proxy-gradient reliability audit that go beyond typical PPG-glucose papers. The fixed-configuration reimplementation of baselines and aggregation-aware reporting are strengths. The clinical and translational significance remains provisional: single-center capillary references, degraded high-glucose performance, and a six-subject pilot limit claims beyond research-stage proof-of-concept, which the authors largely acknowledge.

major comments (4)
  1. Abstract, Table 4, and §4.3 lead with window-level MAE/RMSE/Zone A, but each 40 s acquisition is partitioned into ten 4 s windows that share one capillary BG label (§3.1.2 Stages 4–5; §4.2). Adjacent windows are highly correlated morphology around the same reference, so window-level error can overstate effective sample independence and inflate comparative gains. Tables 5–6 show acquisition- and subject-macro metrics remain favorable (e.g., acquisition MAE 0.60, within-band 93.8%), yet these are secondary. For the central SOTA claim to be load-bearing, Table 4 and the abstract should report acquisition-level (and preferably subject-macro) metrics as co-primary, with window-level retained only as a secondary analysis, and any ranking of methods should be verified at those units.
  2. Table 6 high-glucose slice (y≥10.0 mmol/L): RMSE rises to 2.34 mmol/L and ISO-related within-band falls to 81.7%, versus 1.02 / 93.9% in the 5.55–10 range; Fig. 6 also shows vertical spread near ~16 mmol/L. This is clinically the most consequential regime and is under-emphasized relative to the headline 0.64 MAE and 97.69% A+B. The manuscript should (i) report baseline methods on the same glucose-range slices, (ii) state sample counts per range, and (iii) qualify the main claim so that strong overall averages are not read as uniform accuracy across the 4.2–16.3 mmol/L spectrum.
  3. §3.5.2 Eqs. (12)–(15): ME2AC’s task structure rests on histogram plug-in entropy and mutual information over binned standardized raw PPG-window coordinates. In high-dimensional waveform space this is a coarse discrete morphology proxy; the paper does not show that discovered partitions correspond to interpretable physiological regimes (e.g., perfusion, autonomic state, glycemic context) beyond clustering scores (Tables 10–12). Because DIL’s curriculum, memory allocation m_b, and gradient projection all depend on these partitions, please add evidence that ME2AC tasks are more than a useful regularizer—e.g., stability of partitions under feature alternatives (learned embeddings vs raw bins), or correlation of task identity with metadata available in the benchmark (BP, diabetes status, encounter timing)—or else reframe ME2AC strictly as an operational clustering heuristic rather than latent
  4. Table 4 comparison protocol (§4.3): baselines are reimplemented with original/default settings and no fold-specific tuning, while the proposed method uses a fixed but author-chosen configuration (C_mem=512, α=0.15, 3:1 labeled:unlabeled, ME2AC quantiles, etc.; §4.1). This is preferable to test leakage, but it leaves open whether large margins (e.g., vs Zhang et al. 2024 MAE 0.99) partly reflect unequal optimization effort or backbone capacity rather than DIL/MERDC/PGBA. Please add (i) a controlled backbone-matched comparison (same InceptionTime without DIL modules), already partly in Table 7, elevated into the main comparison narrative, and (ii) at least a minimal sensitivity or published-hyperparameter sweep for the strongest baselines so the ranking is not solely default-vs-tuned.
minor comments (6)
  1. §5.1 / Fig. 9: the seven-day pilot (n=6, alternate wrist device, mixed SMBG/venous references) is useful feasibility evidence but should be more clearly labeled as exploratory and not used to support robustness language that could be read as multi-day clinical validation.
  2. Notation: 𝐉, 𝓁, ̄𝑔, and ̃𝑔 are dense in §3.4–3.6; a short symbol table already exists (Table 1) but does not list all gradient symbols—extend it for readability.
  3. Fig. 10(b): the y-axis is a sampled-window index for visualization only; the caption states this, but the main text still risks being read as a physiological axis—add an explicit one-line caveat in §5.2.
  4. Data availability: controlled-access release is acceptable, but for a paper whose contribution includes establishing a community benchmark, state the access criteria, expected timeline, and what metadata (comorbidities, encounter intervals) will ship with the windows.
  5. Minor consistency: abstract reports 97.69±1.63% A+B while Table 13 mean A+B is 97.70±1.63—align rounding throughout.
  6. §4.5.3 signal-length ablation: only durations near 4 s are discussed in detail; if 1–2 s and >7 s were tested, report the full curve or table values, not only the narrative summary.

Circularity Check

0 steps flagged · score 0.0 of 10

No derivation circularity: performance is measured against held-out capillary references under fold-isolated training; clustering and proxy gradients do not encode the reported errors by construction.

full rationale

The paper’s load-bearing claim is empirical (subject-independent 5CV MAE/RMSE/Clarke on a new PPG–capillary-BG benchmark), not a first-principles derivation that could collapse into its inputs. ME2AC discovers tasks from standardized PPG windows only, without BG labels or test subjects (Eqs. 12–15; §3.5.2; fold isolation in §3.3–3.4 and Fig. 5/Table 3). PGBA is trained by bridge loss on labeled training-fold samples and audited for alignment with supervised gradients; unlabeled windows contribute proxy update directions, not pseudo-labels that would force the reported point errors (Eqs. 17–22; Algorithm 2; Table 8). Hyperparameters are fixed across folds; held-out subjects are used only for frozen inference. Self-citations (e.g., Zhang et al. 2024/2026 for prior static PPG pretraining and backbone choice) motivate the setting and architecture but do not supply a uniqueness theorem or ansatz that forces the headline metrics. Aggregation and high-glucose slice analyses are evaluation-design concerns, not circular reductions of prediction to fit. No step reduces Eq. X to Eq. Y by construction or renames a fitted quantity as an independent prediction.

Assumptions & free parameters 6 free parameters · 6 assumptions · 3 invented entities

The central empirical claim rests on standard supervised/continual-learning math, domain physiology assumptions about PPG–glucose coupling, and several paper-specific modules and hyperparameters that organize tasks and unlabeled updates. No new physical particle or force is postulated; the invented entities are algorithmic constructs whose value is judged by held-out error, not independent physical detection.

free parameters (6)
  • episodic memory capacity C_mem = 512
    Fixed at 512; directly controls replay budget and stability–plasticity trade-off.
  • PGBA scaling factor alpha = 0.15
    Bounds proxy-gradient magnitude relative to supervised gradients; set to 0.15.
  • ME2AC entropy/MI quantiles and bin counts = defaults as in §4.1
    alpha_H=0.60, alpha_MI=0.75, K_H=K_MI=10, k=5 determine task discovery; sensitivity shown but defaults chosen by authors.
  • labeled:unlabeled training-stream ratio = 3:1
    75%/25% split inside training folds withholds labels for PGBA; not learned from first principles.
  • learning rates, batch size, epochs per task = as in §4.1
    lr=1e-3, agent lr=1e-3, batch 128, 20 epochs/task fixed before final 5CV.
  • fold-specific NLL calibration a_k, b_k = per-fold fit on train folds
    Two parameters map raw variance to calibrated log-variance per outer fold for PICP/NLL only.
assumptions (6)
  • domain assumption Within each ME2AC-discovered incremental step, samples are treated as locally IID for empirical risk minimization.
    Stated in §3.3 Eq. (3); required for standard mini-batch updates under non-stationary streams.
  • domain assumption PPG morphology carries a learnable relationship to blood glucose via vascular stiffness, perfusion, and autonomic effects.
    Core physiological premise in Introduction and §5.5; without it the regression target is not identifiable from PPG alone.
  • domain assumption Gradient-episodic-memory non-interference constraints (first-order projection) mitigate catastrophic forgetting for continuous regression tasks.
    Adapted from Lopez-Paz et al. GEM; used in Eqs. (7)–(9).
  • ad hoc to paper Histogram plug-in entropy and mutual information on binned standardized PPG windows are adequate proxies for latent physiological task structure.
    ME2AC design choice in §3.5.2; not a standard theorem that these discrete estimates equal true physiological regimes.
  • domain assumption Capillary Accu-Chek measurements under daily QC are an acceptable reference for longitudinal wearable benchmarking (vs lab venous plasma).
    Explicitly discussed in §5.5; absolute-error interpretation depends on this.
  • standard math Standard calculus / chain-rule gradient updates and Gaussian NLL for uncertainty calibration.
    Eqs. (16)–(22) and (10)–(11).
invented entities (3)
  • Dynamic Incremental Learning (DIL) paradigm for continuous physiological regression
    purpose: Frame continual adaptation under continuous multi-scale drift without class boundaries.
    Paper-defined specialization of incremental learning; evaluated only via this study’s metrics.
  • MERDC / ME2AC mutual-entropy replay clustering
    purpose: Discover meta- and derived-task partitions and allocate replay memory.
    New clustering operator combining adaptive radius, local entropy cores, and MI merging; no external validation outside this pipeline.
  • Proxy Gradient Bridging Agent (PGBA)
    purpose: Map model outputs to bounded output-space proxy gradients so unlabeled PPG can update the regressor without pseudo-labels.
    Paper-specific agent trained by bridge loss; reliability audited only on training-fold labeled batches.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Non-invasive Blood Glucose Estimation from Wearable Physiological Signals." pith.science (2026). https://pith.science/paper/QBM6QGSX

@misc{pith2026260704414,
  author       = {Pith},
  title        = {Pith review of: Non-invasive Blood Glucose Estimation from Wearable Physiological Signals},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QBM6QGSX}},
  note         = {Machine review of arXiv:2607.04414}
}
abstract

Non-invasive blood glucose estimation from wearable physiological signals remains difficult because longitudinal photoplethysmography (PPG) data are subject to distribution drift, whereas reference capillary blood glucose labels are sparse and costly to acquire. We propose a \rev{deep-learning-based} dynamic incremental learning (DIL) framework that combines a mutual entropy-optimized replay-based dynamic clustering module (MERDC) with an uncertainty-quantified proxy gradient bridging agent (PGBA) for label-efficient adaptation to unlabeled PPG streams. To support this setting, we further establish a longitudinal benchmark dataset comprising PPG, reference capillary blood glucose, and cuff blood pressure measurements from 183 participants collected over 285 days, and we make this resource available to the research community. Under 5-fold subject-independent validation, the proposed method achieves a mean absolute error (MAE) of $0.64 \pm 0.01$ millimoles per liter (mmol/L) and a root mean square error (RMSE) of $1.29 \pm 0.10$ mmol/L, with $97.69 \pm 1.63\%$ of estimates falling within Clarke zones A+B. Aggregation-level analyses further support the robustness of the observed error distribution beyond window-level evaluation. \rev{These results provide a proof-of-concept for adaptive non-invasive glucose estimation in wearable physiological sensing and establish a longitudinal benchmark for subsequent research.

Figures

Figures reproduced from arXiv: 2607.04414 by the authors.

Figure 1
Figure 1. Illustrating the technological progression from (a) traditional invasive fingerstick sampling and (b) Continuous Glucose Monitoring (CGM) to (c) our proposed non-invasive method. The feasibility of this approach is established through (d) our systematic data collection protocol, which produced a novel public dataset. Entropy-optimized Enhanced Adaptive Clustering (ME2AC) partitions streaming PPG–glucose data into di… view at source ↗
Figure 2
Figure 2. Overview of the proposed MERDC-enhanced dynamic incremental learning (DIL) framework for non-invasive blood glucose estimation from PPG signals. (a) Stability–plasticity dilemma and unreliable empirical risk minimization under limited labeled data. In the inset, the black curve represents old-data retention associated with stability, whereas the red curve represents new-data adaptation associated with plasticity. (b… view at source ↗
Figure 3
Figure 3. Summary of encounter frequency, temporal distribution, and baseline health profiles in the longitudinal benchmark dataset. (a) Participant distribution by encounter count, stratified by diabetic and non-diabetic status. For groups with repeated encounters, the horizontal interval marker indicates the mean time interval between consecutive encounters, reported in days. (b) Monthly distribution of collected PPG signal… view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Curation pipeline for the study cohort (𝑛 = 195): S1 Data integrity screening removed 3 subjects with invalid (out￾of-range) or log-disconfirmed blood glucose or blood pressure labels. S2 Data availability screening removed 2 subjects whose raw 40 s PPG segments violat…
Figure 5
Figure 5. Figure 5: Leakage-free subject-independent five-fold cross-validation and training-fold MERDC-DIL pipeline. Participants were first partitioned into five mutually exclusive subject folds. In each cycle, one fold was held out for testing and the remaining four folds were used for…
Figure 6
Figure 6. Figure 6: Clarke Error Grid analysis of the proposed MERDC-enhanced dynamic incremental learning framework under subject￾independent evaluation. Each point represents a window-level blood glucose estimate paired with the corresponding reference capillary blood glucose measuremen…
Figure 7
Figure 7. Figure 7: Ablation study of the proposed method, where (a) presents the baseline component analysis and (b) shows the integrated results combining the ablation study outcomes with the evaluation under varying signal lengths [PITH_FULL_IMAGE:figures/full_fig_p022_7.png]
Figure 8
Figure 8. Figure 8: Uncertainty calibration curve under subject-independent 5-fold cross-validation. The dashed diagonal represents perfect agreement between nominal and empirical coverage. Compared with the raw variance output, log-variance calibration improves empirical coverage in the …
Figure 9
Figure 9. Figure 9: Seven-day pilot longitudinal validation across six participants (four with diabetes and two non-diabetic). Our model’s continuous PPG-based glucose prediction (solid lines) is compared with intermittent fingerstick SMBG, where circles represent standard fasting and pos…
Figure 10
Figure 10. Figure 10: Illustration of dynamic adaptation under a non-stationary physiological data stream. (a) Method-wise MAE trajectories across incremental stages. The x-axis denotes the sequential task stage from Task 1 to Task 6, and the colored curves denote comparison methods rather…
Figure 11
Figure 11. Figure 11 [PITH_FULL_IMAGE:figures/full_fig_p030_11.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

51 extracted references · 2 canonical work pages

  1. [1]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume =

    Zhang, Zexing and Lu, Huimin and Zhao, Qingxin , title =. Proceedings of the AAAI Conference on Artificial Intelligence , volume =. 2026 , month =. doi:10.1609/aaai.v40i34.40088 , url =

  2. [2]

    The Lancet , volume=

    Global, regional, and national burden of diabetes from 1990 to 2021, with projections of prevalence to 2050: a systematic analysis for the Global Burden of Disease Study 2021 , author=. The Lancet , volume=. 2023 , publisher=

  3. [3]

    IEEE Transactions on Industrial Electronics , year=

    Liu, Yicun and Zhang, Wan and Liu, Wei and Lu, Yi and Tao, Xueran and Jia, Shiyue and Shi, Dawei , title =. IEEE Transactions on Industrial Electronics , year=

  4. [4]

    and Kuhar, Peter and Hughes, J

    Avram, Robert and Olgin, Jeffrey E. and Kuhar, Peter and Hughes, J. Weston and Marcus, Gregory M. and Pletcher, Mark J. and Aschbacher, Kirstin and Tison, Geoffrey H. , title =. Nature Medicine , volume =. 2020 , publisher =

  5. [5]

    and Ponnapalli, Prasad V

    Ray, Daniel and Collins, Tim and Woolley, Sandra I. and Ponnapalli, Prasad V. S. , title =. IEEE Reviews in Biomedical Engineering , volume =. 2021 , publisher =

  6. [6]

    and Abbott, Derek and Lim, Kenneth and Ward, Rabab , title =

    Elgendi, Mohamed and Fletcher, Richard and Liang, Yongbo and Howard, Newton and Lovell, Nigel H. and Abbott, Derek and Lim, Kenneth and Ward, Rabab , title =. NPJ Digital Medicine , volume =. 2019 , publisher =

  7. [7]

    IEEE Journal of Biomedical and Health Informatics ,year=

    Chen, Shisen and Qin, Fen and Ma, Xuesheng and Wei, Jie and Zhang, Yuan-Ting and Zhang, Yuan and Jovanov, Emil , title =. IEEE Journal of Biomedical and Health Informatics ,year=

  8. [8]

    Journal of Biomedical Informatics , volume =

    Xue, Yuewei and Guan, Shaopeng and Jia, Wanhai , title =. Journal of Biomedical Informatics , volume =. 2024 , publisher =

Show all 51 references
  1. [9]

    IEEE Transactions on Neural Networks and Learning Systems , volume =

    Li, Jingzhen and Ma, Jingjing and Omisore, Olatunji Mumini and Liu, Yuhang and Tang, Huajie and Ao, Pengfei and Yan, Yan and Wang, Lei and Nie, Zedong , title =. IEEE Transactions on Neural Networks and Learning Systems , volume =. 2024 , publisher =

  2. [10]

    IEEE Journal of Biomedical and Health Informatics , volume =

    Ali, Mubashir and Li, Jingzhen and Fan, Bokun and Nie, Zedong , title =. IEEE Journal of Biomedical and Health Informatics , volume =. 2024 , publisher =

  3. [11]

    IEEE Transactions on Industrial Informatics , volume =

    Zhang, Gaobo and Mei, Zhen and Zhang, Yuan and Ma, Xuesheng and Lo, Benny and Chen, Dongyi and Zhang, Yuanting , title =. IEEE Transactions on Industrial Informatics , volume =. 2020 , publisher =

  4. [12]

    Biomedical Signal Processing and Control , volume =

    Nie, Zihan and Rong, Meng and Li, Kaiyang , title =. Biomedical Signal Processing and Control , volume =. 2023 , publisher =

  5. [13]

    IEEE Transactions on Consumer Electronics , volume =

    Wei, Yiting and Liu, Jiaxin and Hu, Lingyue and Ling, Bingo Wing-Kuen and Liu, Qing , title =. IEEE Transactions on Consumer Electronics , volume =. 2023 , publisher =

  6. [14]

    IEEE Journal of Biomedical and Health Informatics , volume =

    Zhang, Chuanhao and Jovanov, Emil and Liao, Hongen and Zhang, Yuan-Ting and Lo, Benny and Zhang, Yuan and Guan, Cuntai , title =. IEEE Journal of Biomedical and Health Informatics , volume =. 2022 , publisher =

  7. [15]

    Biomedical Signal Processing and Control , volume =

    Zhang, Zexing and Lu, Huimin and Ma, Songzhe and Peng, Jianzhong and Lin, Chenglin and Li, Niya and Dong, Bingwang , title =. Biomedical Signal Processing and Control , volume =. 2024 , publisher =

  8. [16]

    IEEE Transactions on Biomedical Circuits and Systems , volume =

    Hina, Aminah and Saadeh, Wala , title =. IEEE Transactions on Biomedical Circuits and Systems , volume =. 2020 , publisher =

  9. [17]

    Sensors , volume =

    Chu, Justin and Yang, Wen-Tse and Lu, Wei-Ru and Chang, Yao-Ting and Hsieh, Tung-Han and Yang, Fu-Liang , title =. Sensors , volume =. 2021 , publisher =

  10. [18]

    Biomedical Signal Processing and Control , volume =

    Prabha, Anju and Yadav, Jyoti and Rani, Asha and Singh, Vijander , title =. Biomedical Signal Processing and Control , volume =. 2022 , publisher =

  11. [19]

    Thirty-Eighth AAAI Conference on Artificial Intelligence,

    Ella Lan , title =. Thirty-Eighth AAAI Conference on Artificial Intelligence,. 2024 , doi =

  12. [20]

    IEEE Transactions on Pattern Analysis and Machine Intelligence ,year=

    Zhou, Da-Wei and Wang, Qi-Wei and Qi, Zhi-Hong and Ye, Han-Jia and Zhan, De-Chuan and Liu, Ziwei , title =. IEEE Transactions on Pattern Analysis and Machine Intelligence ,year=

  13. [21]

    Applied Sciences , volume =

    Islam, Tanvir Tazul and Ahmed, Md Sajid and Hassanuzzaman, Md and Bin Amir, Syed Athar and Rahman, Tanzilur , title =. Applied Sciences , volume =. 2021 , publisher =

  14. [22]

    Advances in Neural Information Processing Systems , year =

    Lopez-Paz, David and Ranzato, Marc'Aurelio , title =. Advances in Neural Information Processing Systems , year =

  15. [23]

    Scientific Reports , volume =

    Lu, Wei-Ru and Yang, Wen-Tse and Chu, Justin and Hsieh, Tung-Han and Yang, Fu-Liang , title =. Scientific Reports , volume =. 2022 , publisher =

  16. [24]

    Frontiers in Digital Health , volume=

    Non invasive blood glucose estimation using green light photoplethysmography and machine learning , author=. Frontiers in Digital Health , volume=

  17. [25]

    and Alex, Zachariah C

    Paul, Brince and Manuel, Melvin P. and Alex, Zachariah C. , title =. 2012 1st International Symposium on Physics and Technology of Sensors (ISPTS-1) , year =

  18. [26]

    2017 , publisher=

    Jendrike, Nina and Baumstark, Annette and Kamecke, Ulrike and Haug, Cornelia and Freckmann, Guido , journal=. 2017 , publisher=

  19. [27]

    , title =

    Hammour, Ghena and Mandic, Danilo P. , title =. Sensors , volume =. 2023 , publisher =

  20. [28]

    IEEE Sensors Journal , volume =

    Lee, Eugene and Lee, Chen-Yi , title =. IEEE Sensors Journal , volume =. 2021 , publisher =

  21. [29]

    2024 IEEE International Symposium on Circuits and Systems (ISCAS) , year =

    Ali, Noor Faris and Aldhaheri, Alyazia and Wodajo, Bethel and Alshamsi, Meera and Alshamsi, Shaikha and Atef, Mohamed , title =. 2024 IEEE International Symposium on Circuits and Systems (ISCAS) , year =

  22. [30]

    IEEE Journal of Biomedical and Health Informatics , volume =

    Ouyang, Chun and Gan, Zhongxue and Zhen, Junjie and Guan, Yuxiang and Zhu, Xing and Zhou, Peng , title =. IEEE Journal of Biomedical and Health Informatics , volume =. 2021 , publisher =

  23. [31]

    Biomedical Signal Processing and Control , volume =

    Explainable deep learning for diabetes diagnosis with. Biomedical Signal Processing and Control , volume =. 2025 , issn =. doi:https://doi.org/10.1016/j.bspc.2024.106902 , author =

  24. [32]

    Cardiovascular Diabetology , volume =

    Sun, Bao and Luo, Zhiying and Zhou, Jiecan , title =. Cardiovascular Diabetology , volume =. 2021 , publisher =

  25. [33]

    and Tuytelaars, Tinne and Tolias, Andreas S

    Van de Ven, Gido M. and Tuytelaars, Tinne and Tolias, Andreas S. , title =. Nature Machine Intelligence , volume =. 2022 , publisher =

  26. [34]

    Neural Networks , volume=

    A survey on few-shot class-incremental learning , author=. Neural Networks , volume=. 2024 , publisher=

  27. [35]

    Silver, David and Huang, Aja and Maddison, Chris J. and Guez, Arthur and Sifre, Laurent and van den Driessche, George and Schrittwieser, Julian and Antonoglou, Ioannis and Panneershelvam, Veda and Lanctot, Marc and Dieleman, Sander and Grewe, Dominik and Kalchbrenner, Nal and ...

  28. [36]

    John M. Jumper and Richard Evans and Alexander Pritzel and Tim Green and Michael Figurnov and Olaf Ronneberger and Kathryn Tunyasuvunakool and Russ Bates and Augustin Zidek and Anna Potapenko and Alex Bridgland and Clemens Meyer and Simon A A Kohl and Andy Ballard and Andrew C...

  29. [37]

    Kudithipudi and M

    D. Kudithipudi and M. Aguilar-Simon and J. Babb and M. Bazhenov and D. Blackiston and J. Bongard and A. P. Brna and S. C. Raja and N. Cheney and J. Clune and A. Daram and S. Fusi and P. Helfer and L. Kay and N. Ketz and Z. Kira and S. Kolouri and J. L. Krichmar and S. Kriegman...

  30. [38]

    Scientific Data , volume =

    Liang, Yongbo and Chen, Zhencheng and Liu, Guiyong and Elgendi, Mohamed , title =. Scientific Data , volume =. 2018 , publisher =

  31. [39]

    and Weber, Jonathan and Webb, Geoffrey I

    Ismail Fawaz, Hassan and Lucas, Benjamin and Forestier, Germain and Pelletier, Charlotte and Schmidt, Daniel F. and Weber, Jonathan and Webb, Geoffrey I. and Idoumghar, Lhassane and Muller, Pierre-Alain and Petitjean, Fran. Data Mining and Knowledge Discovery , volume =. 2020 ...

  32. [40]

    and Cox, Daniel and Gonder-Frederick, Linda A

    Clarke, William L. and Cox, Daniel and Gonder-Frederick, Linda A. and Carter, William and Pohl, Stephen L. , title =. Diabetes Care , volume =. 1987 , publisher =

  33. [41]

    Scientific Reports , volume =

    Zeynali, Mahdi and Alipour, Khalil and Tarvirdizadeh, Bahram and Ghamari, Mohammad , title =. Scientific Reports , volume =. 2025 , publisher =

  34. [42]

    Biomedical Signal Processing and Control , volume =

    Tian, Zhonghe and Liu, Aiping and Zhu, Guokang and Chen, Xun , title =. Biomedical Signal Processing and Control , volume =. 2025 , publisher =

  35. [43]

    Nature Machine Intelligence , volume =

    Hasani, Ramin and Lechner, Mathias and Amini, Alexander and Liebenwein, Lucas and Ray, Aaron and Tschaikowski, Max and Teschl, Gerald and Rus, Daniela , title =. Nature Machine Intelligence , volume =. 2022 , publisher =

  36. [44]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =

    Ma, Xu and Dai, Xiyang and Bai, Yue and Wang, Yizhou and Fu, Yun , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =

  37. [45]

    Pattern Recognition , volume =

    Luo, Yuxiao and Zhang, Songming and Lyu, Ziyu and Hu, Yuhan , title =. Pattern Recognition , volume =. 2025 , publisher =

  38. [46]

    Bioengineering , volume=

    Optimal signal quality index for photoplethysmogram signals , author=. Bioengineering , volume=. 2016 , publisher=

  39. [47]

    IEEE Transactions on Biomedical Engineering , volume=

    Two-stage approach for detection and reduction of motion artifacts in photoplethysmographic data , author=. IEEE Transactions on Biomedical Engineering , volume=. 2010 , publisher=

  40. [48]

    2023 , doi =

    Dongwan Kim and Bohyung Han , title =. 2023 , doi =

  41. [49]

    Nature , volume=

    Loss of plasticity in deep continual learning , author=. Nature , volume=. 2024 , publisher=

  42. [50]

    Frontiers in Public Health , volume=

    Agile human activity recognition for wearable devices based on online incremental learning , author=. Frontiers in Public Health , volume=

  43. [51]

    The Shapley Value in Machine Learning , booktitle =

    Benedek Rozemberczki and Lauren Watson and P. The Shapley Value in Machine Learning , booktitle =. 2022 , doi =

Pith tools

Reviewed July 11, 2026 · model on record in the stance chip above.