Pith. sign in

REVIEW 2 major objections 1 minor 42 references

Learning temporal embeddings from electronic health records of chronic kidney disease patients

T0 review · 2 major / 1 minor · reviewed 2026-05-16 · grok-4.3

Challenge this review Download PDF Read on arXiv ↗

Pith's one-line read Time-aware LSTM models learn more structured embeddings from chronic kidney disease patient records than standard recurrent networks.

desk verdict T-LSTM gives tighter CKD stage clusters and embedding models beat end-to-end on mortality in this MIMIC-IV slice, but missing split details leave leakage risk unaddressed. read the letter →

arxiv 2601.18675 v2 submitted 2026-01-26 cs.LG cs.AI

Aditya Kumar , Mario A. Cypko , Oliver Amft This is my paper
classification cs.LGcs.AI
keywords temporalembeddingselectronichealthrecordschronickidneydiseaseT-LSTMrepresentationlearningclinicalpredictionMIMIC-IV
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

The paper tests whether recurrent architectures can extract clinically useful temporal patterns from electronic health records without tying the model to one specific prediction task. It trains vanilla LSTM, attention-augmented LSTM, and time-aware LSTM networks on longitudinal data from chronic kidney disease patients, then evaluates the resulting embeddings on disease-stage clustering and mortality prediction. The time-aware variant produces tighter patient groupings and, when used as an intermediate representation, raises downstream accuracy above what end-to-end training achieves on the same data.

What carries the argument

The T-LSTM, a recurrent unit that modifies standard LSTM gates to incorporate explicit time intervals between clinical observations, allowing it to model irregular sampling common in electronic health records.

What would settle it

Re-training the same three architectures on a different longitudinal cohort and observing that T-LSTM embeddings no longer show lower Davies-Bouldin index or higher downstream accuracy would indicate the advantage is not general.

Watch

Extended reading notes

Core claim

Training a time-aware LSTM on longitudinal records yields embeddings that achieve a Davies-Bouldin index of 9.91 and CKD stage classification accuracy of 0.74, outperforming both vanilla LSTM and attention-augmented LSTM. These embeddings also raise in-ICU mortality prediction accuracy to 0.82-0.83 when supplied to a downstream classifier, exceeding the performance of models trained end-to-end directly on the raw sequences.

Load-bearing premise

That performance on CKD stage clustering and ICU mortality within the MIMIC-IV cohort reflects general embedding quality rather than task-specific or dataset-specific artifacts.

Editorial extensions

If this is right

  • T-LSTM embeddings group patients by CKD stage more cleanly than embeddings from vanilla or attention LSTMs.
  • Learning embeddings as an intermediate step improves mortality prediction accuracy over direct end-to-end training on the same records.
  • Recurrent models that respect time intervals between visits capture temporal structure in observational clinical data more effectively.
  • Task-agnostic embeddings support multiple downstream clinical predictions without retraining the full model each time.

Reading between the lines

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

  • The same embedding approach could be tested on other chronic conditions whose records also arrive at irregular intervals.
  • Hospitals with differing visit frequencies might obtain similar gains by adding explicit time handling to their recurrent models.
  • More structured embeddings could reduce the amount of labeled data needed when adapting a model to a new clinical site.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 1 minor

Summary. The paper investigates whether temporal embedding models trained on longitudinal electronic health records from MIMIC-IV CKD patients can learn clinically meaningful representations. It compares vanilla LSTM, attention-augmented LSTM, and time-aware LSTM (T-LSTM) architectures, both as embedding models and end-to-end predictors. The central claims are that T-LSTM yields more structured embeddings (DBI=9.91, CKD stage classification accuracy=0.74) than the baselines (DBI=15.85/20.72, accuracies=0.63/0.67), and that all embedding models outperform direct end-to-end predictors on in-ICU mortality prediction (accuracy 0.82-0.83 vs. 0.72-0.75).

Significance. If the results hold under rigorous validation, the work provides evidence that intermediate representation learning with recurrent architectures can capture temporal disease dynamics in EHR data more effectively than task-specific end-to-end training, supporting the development of generalizable embeddings for model-guided medicine. The evaluation via external downstream tasks (clustering and mortality) is a positive design choice that avoids direct circularity with the training objective.

major comments (2)
  1. Abstract: The reported metrics (DBI values and accuracies) are presented without any information on train/test splits, patient-disjoint partitioning, hyperparameter search, or statistical significance testing. This omission is load-bearing because the headline superiority of T-LSTM embeddings and the embedding-vs-end-to-end gains cannot be verified without confirming that downstream evaluation data did not leak into embedding training.
  2. Abstract: No description is given of how irregular sampling, missing values, or variable-length sequences in the MIMIC-IV CKD cohort are handled within the T-LSTM (or baseline) architectures. Since the central claim concerns temporal structure capture, the absence of these implementation details prevents assessment of whether the DBI and accuracy improvements genuinely reflect better modeling of disease dynamics.
minor comments (1)
  1. Abstract: The abstract would be clearer if it briefly stated the cohort size (number of patients and records) and the exact clustering procedure used for the Davies-Bouldin Index evaluation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for these constructive comments on methodological transparency. Both points identify genuine gaps in the abstract that could affect verifiability of the central claims. We have revised the abstract to include concise statements on data partitioning, evaluation protocol, and handling of EHR irregularities, while preserving the original length constraints. The full details remain in the Methods section.

read point-by-point responses

  1. Referee: Abstract: The reported metrics (DBI values and accuracies) are presented without any information on train/test splits, patient-disjoint partitioning, hyperparameter search, or statistical significance testing. This omission is load-bearing because the headline superiority of T-LSTM embeddings and the embedding-vs-end-to-end gains cannot be verified without confirming that downstream evaluation data did not leak into embedding training.

    Authors: We agree that these details are essential. The Methods section (3.1) describes a patient-level disjoint split (70% train, 15% validation, 15% test) with no patient overlap between embedding training and downstream evaluation sets. Hyperparameters were tuned via grid search on the validation set; all metrics are reported as mean ± std over five independent runs with different seeds. We have added the following sentence to the abstract: 'All experiments use patient-disjoint 70/15/15 splits with hyperparameters selected on a held-out validation set and results averaged over five runs.' No formal statistical significance tests (e.g., paired t-tests) were conducted beyond reporting variability; this is a limitation we acknowledge and will address in revisions if space permits. revision: yes

  2. Referee: Abstract: No description is given of how irregular sampling, missing values, or variable-length sequences in the MIMIC-IV CKD cohort are handled within the T-LSTM (or baseline) architectures. Since the central claim concerns temporal structure capture, the absence of these implementation details prevents assessment of whether the DBI and accuracy improvements genuinely reflect better modeling of disease dynamics.

    Authors: The T-LSTM architecture explicitly incorporates inter-visit time deltas to modulate cell-state updates, directly addressing irregular sampling. Missing values are forward-filled for longitudinal features and imputed with cohort mode for static variables; variable-length sequences are zero-padded with a binary mask that excludes padded timesteps from loss and gradient computation. These procedures are detailed in Section 3.2. To address the abstract gap we have inserted: 'T-LSTM handles irregular intervals via explicit time-delta inputs, with forward imputation for missing values and masking for variable-length sequences.' This addition clarifies that the reported gains arise from the model's temporal modeling capacity. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity detected in derivation or evaluation chain

full rationale

The paper trains three recurrent architectures (vanilla LSTM, attention-augmented LSTM, T-LSTM) on MIMIC-IV CKD patient sequences both as embedding generators and as direct end-to-end predictors. Embedding quality is assessed via two external downstream tasks—CKD stage clustering (DBI and classification accuracy) and in-ICU mortality prediction—whose metrics are not algebraically defined by the training loss or fitted parameters. No equations are presented that reduce reported gains (e.g., DBI 9.91 or accuracy 0.82–0.83) to quantities already fixed by construction. No self-citations, uniqueness theorems, or ansatzes are invoked to justify architectural choices or performance claims. The evaluation therefore remains independent of the training objective, yielding a self-contained empirical comparison.

Assumptions & free parameters 1 free parameters · 2 assumptions · 0 invented entities

The central claims rest on standard neural-network assumptions (gradient descent finds useful representations, recurrent models capture temporal dependencies in irregular EHR) plus dataset-specific choices (CKD cohort definition, feature extraction from MIMIC-IV). No new entities are postulated. Hyperparameters such as hidden size, learning rate, and sequence length are free parameters fitted during training but not enumerated in the abstract.

free parameters (1)
  • model hyperparameters (hidden size, learning rate, dropout)
    Chosen during training to optimize the reported metrics; exact values not stated in abstract.
assumptions (2)
  • domain assumption Recurrent architectures can model temporal structure in observational clinical data
    Invoked in the motivation for using LSTM variants on longitudinal EHR.
  • domain assumption Davies-Bouldin Index and downstream classifier accuracy are valid proxies for embedding quality
    Used to evaluate the learned representations without further justification in the abstract.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning temporal embeddings from electronic health records of chronic kidney disease patients." pith.science (2026). https://pith.science/paper/2601.18675

@misc{pith2026260118675,
  author       = {Pith},
  title        = {Pith review of: Learning temporal embeddings from electronic health records of chronic kidney disease patients},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2601.18675}},
  note         = {Machine review of arXiv:2601.18675}
}
read the original abstract

We investigate whether temporal embedding models trained on longitudinal electronic health records can learn clinically meaningful representations without compromising predictive performance, and how architectural choices affect embedding quality. Model-guided medicine requires representations that capture disease dynamics while remaining transparent and task agnostic, whereas most clinical prediction models are optimised for a single task. Representation learning facilitates learning embeddings that generalise across downstream tasks, and recurrent architectures are well-suited for modelling temporal structure in observational clinical data. Using the MIMIC-IV dataset, we study patients with chronic kidney disease (CKD) and compare three recurrent architectures: a vanilla LSTM, an attention-augmented LSTM, and a time-aware LSTM (T-LSTM). All models are trained both as embedding models and as direct end-to-end predictors. Embedding quality is evaluated via CKD stage clustering and in-ICU mortality prediction. The T-LSTM produces more structured embeddings, achieving a lower Davies-Bouldin Index (DBI = 9.91) and higher CKD stage classification accuracy (0.74) than the vanilla LSTM (DBI = 15.85, accuracy = 0.63) and attention-augmented LSTM (DBI = 20.72, accuracy = 0.67). For in-ICU mortality prediction, embedding models consistently outperform end-to-end predictors, improving accuracy from 0.72-0.75 to 0.82-0.83, which indicates that learning embeddings as an intermediate step is more effective than direct end-to-end learning.

Figures

Figures reproduced from arXiv: 2601.18675 by the authors.

Figure 1
Figure 1. Overview of the approach. (a) Longitudinal EHR data are extracted and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Recurrent model architectures evaluated in this study. (a) Vanilla [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. t-SNE visualisation of embedding spaces separated by classes 0-7 based on ICD codes for CKD stages. (a) Embedding space of vanilla LSTM. (b) [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Sign in to comment.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages

  1. [1]

    and Golubnitschaja, O., 2014

    Lemke, H.U. and Golubnitschaja, O., 2014. Towards personal health care with model-guided medicine: long-term PPPM-related strategies and realisation opportunities within ‘Horizon 2020’. EPMA Journal, 5(1), p.8

    work page 2014

  2. [2]

    and Wilhelm, D., 2024

    Cypko, M.A. and Wilhelm, D., 2024. Ladies and Gentlemen! This is no humbug. Why Model-Guided Medicine will become a main pillar for the future healthcare system. International Journal of Computer Assisted Radiology and Surgery, 19(10), pp.1919-1927

    work page 2024

  3. [3]

    and Tawhai, M.,

    Chase, J.G., Preiser, J.C., Dickson, J.L., Pironet, A., Chiew, Y .S., Pretty, C.G., Shaw, G.M., Benyo, B., Moeller, K., Safaei, S. and Tawhai, M.,

    work page

  4. [4]

    Biomedical engineering online, 17(1), p.24

    Next-generation, personalised, model-based critical care medicine: a state-of-the art review of in silico virtual patient models, methods, and cohorts, and how to validation them. Biomedical engineering online, 17(1), p.24

    work page

  5. [5]

    Envisioning the future of personalized medicine: role and realities of digital twins

    Vall ´ee, A., 2024. Envisioning the future of personalized medicine: role and realities of digital twins. Journal of medical Internet research, 26, p.e50204

    work page 2024

  6. [6]

    and Subasi, M.E., 2024

    Subasi, A. and Subasi, M.E., 2024. Digital twins in healthcare and biomedicine. In Artificial Intelligence, Big Data, Blockchain and 5G for the Digital Transformation of the Healthcare Industry (pp. 365-401). Academic Press

    work page 2024

  7. [7]

    and Benson, M., 2025

    Li, X., Loscalzo, J., Mahmud, A.F., Aly, D.M., Rzhetsky, A., Zitnik, M. and Benson, M., 2025. Digital twins as global learning health and disease models for preventive and personalized medicine. Genome Medicine, 17(1), p.11

    work page 2025

  8. [8]

    and Kaur, S., 2025

    Sharma, H. and Kaur, S., 2025. Patient-specific digital twins for person- alized healthcare: a hybrid AI and simulation-based framework. IEEE Access

    work page 2025

Show all 42 references
  1. [9]

    and Umadevi, K.S., 2025

    Iyer, A.A. and Umadevi, K.S., 2025. Design and analysis of TwinCardio framework to detect and monitor cardiovascular diseases using digital twin and deep neural network. Scientific Reports, 15(1), p.24376

    2025

  2. [10]

    and Salimi-Khorshidi, G., 2020

    Li, Y ., Rao, S., Solares, J.R.A., Hassaine, A., Ramakrishnan, R., Canoy, D., Zhu, Y ., Rahimi, K. and Salimi-Khorshidi, G., 2020. BEHRT: trans- former for electronic health records. Scientific reports, 10(1), p.7155

    2020

  3. [11]

    and Gao, J., 2017, August

    Ma, F., Chitta, R., Zhou, J., You, Q., Sun, T. and Gao, J., 2017, August. Dipole: Diagnosis prediction in healthcare via attention-based bidirectional recurrent neural networks. In Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mini...

    2017

  4. [12]

    and Zhang, P., 2020

    Zhang, D., Yin, C., Zeng, J., Yuan, X. and Zhang, P., 2020. Combining structured and unstructured data for predictive models: a deep learning approach. BMC medical informatics and decision making, 20(1), p.280

    2020

  5. [13]

    and Roberts, K., 2021

    Si, Y ., Du, J., Li, Z., Jiang, X., Miller, T., Wang, F., Zheng, W.J. and Roberts, K., 2021. Deep representation learning of patient data from Electronic Health Records (EHR): A systematic review. Journal of biomedical informatics, 115, p.103671

    2021

  6. [14]

    and Fiterau, M., 2021, August

    Deznabi, I., Iyyer, M. and Fiterau, M., 2021, August. Predicting in- hospital mortality by combining clinical notes with time-series data. In Findings of the association for computational linguistics: ACL-IJCNLP 2021 (pp. 4026-4031)

    2021

  7. [15]

    and Martini, A., 2025

    Shivashankar, K., Hajj, G.S.A. and Martini, A., 2025. Scalability and Maintainability Challenges and Solutions in Machine Learning: Sys- tematic Literature Review. arXiv preprint arXiv:2504.11079

    2025

  8. [16]

    Learning task-state representations

    Niv, Y ., 2019. Learning task-state representations. Nature neuroscience, 22(10), pp.1544-1553

    2019

  9. [17]

    and Glicksberg, B.S., 2025

    Kauffman, J., Miotto, R., Klang, E., Costa, A., Norgeot, B., Zitnik, M., Khader, S., Wang, F., Nadkarni, G.N. and Glicksberg, B.S., 2025. Embedding Methods for Electronic Health Record Research. Annual Review of Biomedical Data Science, 8

    2025

  10. [18]

    Machine Learning for Complex Clinical Time Series: From Representation Learning to Interpretable Phenotyping

    Zheng, X., 2025. Machine Learning for Complex Clinical Time Series: From Representation Learning to Interpretable Phenotyping

    2025

  11. [19]

    and Gao, J.,

    Ruan, T., Lei, L., Zhou, Y ., Zhai, J., Zhang, L., He, P. and Gao, J.,

  12. [20]

    BMC medical informatics and decision making, 19(Suppl 8), p.259

    Representation learning for clinical time series prediction tasks in electronic health records. BMC medical informatics and decision making, 19(Suppl 8), p.259

  13. [21]

    and Liu, N., 2022

    Xie, F., Yuan, H., Ning, Y ., Ong, M.E.H., Feng, M., Hsu, W., Chakraborty, B. and Liu, N., 2022. Deep learning for temporal data representation in electronic health records: A systematic review of challenges and methodologies. Journal of biomedical informatics, 126, p.103980

    2022

  14. [22]

    and Chiang, J.N., 2025

    [16] Lee, S.A., Jain, S., Chen, A., Ono, K., Biswas, A., Rudas, ´A., Fang, J. and Chiang, J.N., 2025. Clinical decision support using pseudo-notes from multiple streams of EHR data. npj Digital Medicine, 8(1), p.394

    2025

  15. [23]

    and Hu, G., 2025

    Shen, Y ., Yu, J., Zhou, J. and Hu, G., 2025. Twenty-five years of evolution and hurdles in electronic health records and interoperability in medical research: comprehensive review. Journal of Medical Internet Research, 27, p.e59024

    2025

  16. [24]

    and Huang, S., 2025

    Zheng, Z., Luo, J., Zhu, Y ., Du, L., Lan, L., Zhou, X., Yang, X. and Huang, S., 2025. Development and Validation of a Dynamic Real- Time Risk Prediction Model for Intensive Care Units Patients Based on Longitudinal Irregular Data: Multicenter Retrospective Study. Journal of m...

    2025

  17. [25]

    and Fan, X., 2025

    Zhong, S., Wang, L.R., Zhan, Z., Ng, Y .Y . and Fan, X., 2025. A Hybrid Approach for Irregular-Time Series Prediction using Electronic Health Records: an Intensive Care Unit Mortality Case Study. ACM Transactions on Computing for Healthcare, 6(4), pp.1-33

    2025

  18. [26]

    and Honavar, V ., 2025

    Ren, W., Zhu, J., Liu, Z., Zhao, T. and Honavar, V ., 2025. A compre- hensive survey of electronic health record modeling: From deep learning approaches to large language models. arXiv preprint arXiv:2507.12774

    2025

  19. [27]

    and Karim, M., 2019, May

    Islam, M.S., Umran, H.M., Umran, S.M. and Karim, M., 2019, May. Intelligent healthcare platform: cardiovascular disease risk factors pre- diction using attention module based LSTM. In 2019 2nd international conference on artificial intelligence and big data (ICAIBD) (pp. 167- ...

    2019

  20. [28]

    and Zhou, J., 2017, August

    Baytas, I.M., Xiao, C., Zhang, X., Wang, F., Jain, A.K. and Zhou, J., 2017, August. Patient subtyping via time-aware LSTM networks. In Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining (pp. 65-74)

    2017

  21. [29]

    and Lovis, C., 2025

    Zheng, Y ., Bensahla, A., Bjelogrlic, M., Zaghir, J., Turbe, H., Bed- narczyk, L., Gaudet-Blavignac, C., Ehrsam, J., Marchand-Maillet, S. and Lovis, C., 2025. A scoping review of self-supervised representation learning for clinical decision making using EHR categorical data. N...

    2025

  22. [30]

    and Lehman, L.W.H., 2023

    Johnson, A.E., Bulgarelli, L., Shen, L., Gayles, A., Shammout, A., Horng, S., Pollard, T.J., Hao, S., Moody, B., Gow, B. and Lehman, L.W.H., 2023. MIMIC-IV , a freely accessible electronic health record dataset. Scientific data, 10(1), p.1

    2023

  23. [31]

    and Beheshti, R., 2022, November

    Gupta, M., Gallamoza, B., Cutrona, N., Dhakal, P., Poulain, R. and Beheshti, R., 2022, November. An extensive data processing pipeline for mimic-iv. In Machine learning for health(pp. 311-325). PMLR

    2022

  24. [32]

    and Fockler, T.V ., 2016

    Stephens, J.H., Ledlow, G.R. and Fockler, T.V ., 2016. Converting ICD-9 to ICD-10. Hospital topics, 94(1), pp.1-7

    2016

  25. [33]

    and Klar, R., 1996

    Zaiss, A., Schulz, S., Graubner, B. and Klar, R., 1996. Conversion Table between ICD-9 and ICD-10. In Medical Informatics Europe’96 (pp. 193- 197). IOS Press

    1996

  26. [34]

    and Jurman, G., 2025

    Chicco, D., Campagner, A., Spagnolo, A., Ciucci, D. and Jurman, G., 2025. The Silhouette coefficient and the Davies-Bouldin index are more informative than Dunn index, Calinski-Harabasz index, Shannon entropy, and Gap statistic for unsupervised clustering internal evaluation o...

    2025

  27. [35]

    and Heo, J.P., 2023

    Kim, J., Lee, M. and Heo, J.P., 2023. Self-feedback detr for temporal action detection. In Proceedings of the IEEE/CVF International Confer- ence on Computer Vision (pp. 10286-10296)

    2023

  28. [36]

    and Yu, G.I., 2025

    Sim, T., Hahn, S., Kim, K.J., Cho, E.Y ., Jeong, Y ., Kim, J.H., Ha, E.Y ., Kim, I.C., Park, S.H., Cho, C.H. and Yu, G.I., 2025. Preserving informative presence: How missing data and imputation strategies affect the performance of an AI-based early warning score. Journal of Cl...

    2025

  29. [37]

    and Dong Thi Hoai, T., 2023

    Hernandez, B., Stiff, O., Ming, D.K., Ho Quang, C., Nguyen Lam, V ., Nguyen Minh, T., Nguyen Van Vinh, C., Nguyen Minh, N., Nguyen Quang, H., Phung Khanh, L. and Dong Thi Hoai, T., 2023. Learning meaningful latent space representations for patient risk stratification: Model de...

    2023

  30. [38]

    and Hospedales, T.M., 2022

    Ericsson, L., Gouk, H., Loy, C.C. and Hospedales, T.M., 2022. Self- supervised representation learning: Introduction, advances, and chal- lenges. IEEE Signal Processing Magazine, 39(3), pp.42-62

    2022

  31. [39]

    and Rajpurkar, P., 2023

    Moor, M., Banerjee, O., Abad, Z.S.H., Krumholz, H.M., Leskovec, J., Topol, E.J. and Rajpurkar, P., 2023. Foundation models for generalist medical artificial intelligence. Nature, 616(7956), pp.259-265

    2023

  32. [40]

    and Zhang, D., 2025

    Zheng, C., Zhu, Q., Fei, L., Li, S., Zhai, X.B., Zhang, D. and Zhang, D., 2025. Disentangled Representation Learning for Robust Brainprint Recognition. IEEE Transactions on Information Forensics and Security

    2025

  33. [41]

    and Hu, D., 2025

    Chen, H., Sun, J., Liu, Y ., Shi, H. and Hu, D., 2025. Self-supervised disentangled representation learning via compositional invariance. IEEE Transactions on Circuits and Systems for Video Technology

    2025

  34. [42]

    and Steyerberg, E.W., 2019

    Van Calster, B., McLernon, D.J., Van Smeden, M., Wynants, L. and Steyerberg, E.W., 2019. Calibration: the Achilles heel of predictive analytics. BMC medicine, 17(1), p.230

    2019

Pith tools

Reviewed May 16, 2026 · model on record in the stance chip above.