Description This record accompanies a manuscript validating a single-token, on-device encoder for continuous glucose monitoring (CGM). Each glucose window is reduced to one compact learned token from which a decision is read on the sensor itself. Under strict pre-registration — frozen hypotheses, patient-disjoint splits, and bootstrap confidence intervals — the encoder predicts hypoglycemic and hyperglycemic excursions 30–60 minutes ahead (AUC 0.93–0.97) and, trained on a single cohort, generalizes without retraining to nine independent public CGM datasets (mean AUC 0.878). A population-level federated refresh adds a further measured gain, while per-individual personalization and multi-signal fusion are shown, with honest boundaries, to add little. Method companion: Paper 19 (doi:10.5281/zenodo.20788187). Keywords: continuous glucose monitoring; hypoglycemia prediction; hyperglycemia prediction; class-discriminant codebook; vector quantization; on-device machine learning; edge inference; cross-dataset generalization; federated refresh; pre-registration; type 1 diabetes; type 2 diabetes References 1. N. Tishby, F. C. Pereira, and W. Bialek, "The information bottleneck method," in Proc. 37th Allerton Conf. Communication, Control, and Computing, 1999, pp. 368–377. 2. R. M. Gray, "Vector quantization," IEEE ASSP Magazine, vol. 1, no. 2, pp. 4–29, 1984. 3. R. A. Fisher, "The use of multiple measurements in taxonomic problems," Annals of Eugenics, vol. 7, no. 2, pp. 179–188, 1936. 4. S. P. Lloyd, "Least squares quantization in PCM," IEEE Transactions on Information Theory, vol. 28, no. 2, pp. 129–137, 1982. 5. Q. Zhao et al., "Chinese diabetes datasets for data-driven machine learning," Scientific Data, vol. 10, no. 35, 2023. 6. J. I. Hidalgo, J. Alvarado, M. Botella, A. Aramendi, J. M. Velasco, and O. Garnica, "HUPA-UCM diabetes dataset," Data in Brief, vol. 55, art. 110559, 2024, doi:10.1016/j.dib.2024.110559. 7. T. Battelino et al., "Clinical targets for continuous glucose monitoring data interpretation: recommendations from the international consensus on time in range," Diabetes Care, vol. 42, no. 8, pp. 1593–1603, 2019. 8. S. Oviedo, J. Vehí, R. Calm, and J. Armengol, "A review of personalized blood glucose prediction strategies for T1DM patients," International Journal for Numerical Methods in Biomedical Engineering, vol. 33, no. 6, e2833, 2017. 9. B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, "Communication-efficient learning of deep networks from decentralized data," in Proc. 20th Int. Conf. Artificial Intelligence and Statistics (AISTATS), 2017, pp. 1273–1282. 10. B. A. Nosek, C. R. Ebersole, A. C. DeHaven, and D. T. Mellor, "The preregistration revolution," Proc. National Academy of Sciences, vol. 115, no. 11, pp. 2600–2606, 2018. 11. B. Efron and R. J. Tibshirani, An Introduction to the Bootstrap. New York: Chapman 64/084,807; 64/084,817; 64/084,821); patent rights are separate from the copyright license. Public datasets (figshare 20444397, CC BY 4.0; HuggingFace byluuu/gluco-tsfm-benchmark) retain their own licenses and are not redistributed in the archive. Licensing inquiries: randolphf@fieldstoneanalyticsllc.com. Randolph James Ferlic, M.D. · Kimberly Kate Ferlic · Fieldstone Analytics, LLC · DOI 10.5281/zenodo.21114273
Ferlic et al. (Wed,) studied this question.