Pith. sign in

REVIEW 3 major objections 5 minor 34 references

Using the CNS–ANS split as a physiological prior, hierarchical contrastive learning plus long-horizon latent prediction yields more transferable sleep representations than topology-agnostic fusion of PSG signals.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-13 07:01 UTC pith:U3ZHFKEZ

load-bearing objection Solid sleep foundation-model paper with real OOD/missing-modality gains; the CNS/ANS hierarchy is a clean packaging idea, but the causal claim that the physiological prior drives the wins is not isolated from scale and architecture. the 3 major comments →

arxiv 2607.07720 v2 pith:U3ZHFKEZ submitted 2026-07-04 cs.LG cs.AI

Omni-Sleep: A Sleep Foundation Model via Hierarchical Contrastive Learning of CNS-ANS Dynamics

Zhoujie Hou , Song Wang , Kexin Lou , Mo Wang , Chen Wei , Quanying Liu This is my paper
classification cs.LG cs.AI
keywords Sleepself-supervised learningmultimodal PSGfoundation modelCNS-ANS couplingcontrastive learningsleep stagingmissing modalities
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Sleep physiology is coordinated brain–body activity: central neural signals (EEG, EOG, EMG) and autonomic cardio-respiratory signals (ECG, respiration) recorded together in polysomnography. Many sleep foundation models still fuse those channels in a flat shared space and ignore that organization. Omni-Sleep treats the CNS/ANS partition as an inductive prior. It trains for agreement among modalities inside each subsystem, alignment between the two subsystem summaries, and latent prediction over long multi-epoch windows that capture night-scale dynamics. Pretrained on more than 100,000 hours of multi-center PSG, the model improves out-of-domain sleep staging and multi-disease classification, works with fewer labels, and stays useful when channels are missing. The paper’s claim is that physiological hierarchy is a practical prior for generalizable sleep representation learning.

Core claim

Omni-Sleep establishes that topology-constrained pretraining built on a CNS/ANS partition—intra-system consistency, inter-system synchronization, and long-horizon latent masked modeling—learns sleep representations that outperform flat-fusion foundation-model baselines on sleep staging and multi-label disease classification, with better label efficiency, cross-dataset transfer, and performance under missing modalities.

What carries the argument

Hierarchical contrastive learning under a CNS/ANS topology: each modality is pulled toward the mean of the other modalities in its subsystem (intra-system InfoNCE), pooled CNS and ANS summaries are aligned with symmetric InfoNCE (inter-system), and latent-space masked prediction over long windows captures macro-scale sleep dynamics.

Load-bearing premise

The hard split of channels into CNS versus ANS groups, and building positives as the average of the other channels in the same group, is the load-bearing prior—not mainly scale, capacity, or generic multi-view contrastive learning.

What would settle it

Train a same-capacity twin with the same encoders and long-horizon latent masking, but replace hierarchical CNS/ANS contrastive losses with a single flat multi-modal InfoNCE over all channels; if out-of-domain staging and multi-disease AUROC match Omni-Sleep under identical linear probing, the physiological topology prior is not what carries the claimed gains.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Downstream sleep staging can approach full-supervision performance with far fewer expert labels after this pretraining.
  • Linear probing on held-out cohorts can transfer without task-specific backbone updates.
  • Staging remains usable under restricted channel sets (EEG-only, CNS-only, ANS-only, or full PSG).
  • Multi-disease linear probes improve for sleep, respiratory, and cardiovascular outcomes when long-horizon modeling is kept.
  • PSG foundation models should treat physiological organization as a first-class constraint rather than generic multimodal fusion.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same subsystem hierarchy prior may help other brain–body monitoring settings where flat fusion also blurs neural versus cardio-respiratory structure.
  • If mean-of-other-modalities positives drive missing-channel behavior, simpler multi-view consistency without an explicit CNS/ANS label set might recover part of the gain.
  • Because coupling is stage-dependent, conditioning inter-system alignment on stage or circadian phase is a natural next test.
  • Subsystem embeddings could support separate neural and autonomic risk heads while still sharing one pretrained backbone.

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

3 major / 5 minor

Summary. Omni-Sleep is a multimodal sleep foundation model that treats the CNS/ANS partition of PSG modalities (EEG/EOG/EMG vs ECG/respiration) as a physiological prior for topology-constrained pretraining. It combines three objectives: micro-scale intra-system InfoNCE with mean-of-other-modalities positives within each subsystem (Eqs. 4–5), macro-scale symmetric inter-system alignment of pooled CNS and ANS summaries (Eqs. 6–7), and long-horizon latent masked prediction with L1 loss over multi-epoch windows (Eqs. 9–11). Pretrained on >100k hours of multi-center PSG (SHHS, WSC, MESA), the model is evaluated via linear probing and few-shot fine-tuning on OOD sleep staging (CinC, ISRUC) and multi-label disease classification (SHHS holdout), reporting gains over SleepFM and SleepGPT under full and missing-modality settings, plus improved label efficiency.

Significance. If the hierarchical CNS–ANS inductive bias is genuinely responsible for the reported gains, the work would be a useful step beyond flat multimodal fusion in sleep foundation models, with practical relevance for label-scarce and incomplete-sensor clinical PSG. Strengths include large multi-center pretraining scale, explicit missing-modality evaluation (Table 1), external cohorts (CinC, ISRUC), multi-disease linear probing (Table 2), and public code. The contribution is currently best read as a strong empirical system paper; the causal claim that physiological hierarchy—not capacity, multi-view contrastive learning, or latent temporal modeling—drives the gains remains only partially supported and is the main limit on significance.

major comments (3)
  1. The central claim that the CNS/ANS hierarchical prior drives gains is under-isolated. §3.1–3.3 introduce the hard partition and mean-of-other positives (Eqs. 4–7) as the key inductive bias, but the only ablation (Omni-Sleep*, Table 2) removes the RoFormer temporal module, not the topology. There is no same-architecture control that replaces hierarchical positives with flat multi-view InfoNCE (or shuffled partitions) while holding data, schedule, and L_p fixed. Without that, Table 1 OOD staging and missing-modality gains can be explained by capacity (56M vs SleepFM 4.4M), multi-view contrastive learning, or latent MAE rather than physiological hierarchy.
  2. Baseline comparisons confound architecture and scale. §4.1 compares Omni-Sleep (56M) to SleepFM (4.4M) and SleepGPT (134M) without matched-capacity or matched-pretraining-recipe controls. Table 1 linear-probing improvements are consistent across EEG/CNS/ANS/Full, but cannot cleanly attribute superiority to the hierarchical objective versus encoder design, pretraining corpus, or optimization. A capacity-matched flat-fusion Omni-Sleep variant, or at least reporting of pretraining compute and data overlap with baselines, is needed for the SOTA claim in §1.
  3. Disease-classification evidence is uneven and partially weakens the long-horizon story. In Table 2, Omni-Sleep improves several AUROCs (e.g., OSA 0.841, COPD/emphysema 0.750, heart failure 0.825), but clinical depression is lower than Omni-Sleep* (0.519 vs 0.545) and hypersomnia remains modest with large variance. The text asserts that long-horizon modeling helps respiratory/cardiovascular outcomes, yet the ablation is only RoFormer removal and does not isolate L_p (Eq. 11) from L_c. Stronger claims about multi-disease utility should be tempered or supported by objective-level ablations and multiple-testing-aware reporting.
minor comments (5)
  1. §3.5 states a two-stage schedule (contrastive warm-up then joint L_c + L_p) but does not report α, β, λ, τ, mask ratio, or long-window length L used in the final model; these free parameters should be listed for reproducibility.
  2. Figure 3 (few-shot ISRUC) is referenced without numerical table values or confidence intervals; adding exact Macro-F1/κ at each label fraction would strengthen the label-efficiency claim.
  3. Notation mixes H^(m), S^(m), s^(m), h_t^(m) for related embeddings; a short glossary or consistent casing would reduce ambiguity in §3.2–3.4.
  4. Related Work cites brain–heart coupling well, but a brief comparison to other topology-aware multimodal SSL (beyond sleep) would better situate the hierarchical contrastive design.
  5. Minor typos/formatting: “brain–body” vs “brain-body”, arXiv header date style, and occasional missing spaces in compound terms (e.g., “topology-agnosticmanner” in the abstract source).

Circularity Check

0 steps flagged

No significant circularity: self-supervised pretraining objectives are independent of downstream labels and metrics; gains are empirical, not definitional.

full rationale

Omni-Sleep is an empirical foundation-model paper. The CNS/ANS partition (M_C vs M_A) and hierarchical InfoNCE construction (Eqs. 4–8) plus latent L1 masked prediction (Eqs. 9–11) are design choices and training losses; they do not algebraically restate sleep-stage F1/κ or disease AUROC. Pretraining is label-free on multi-center PSG; evaluation uses held-out SHHS subjects and external CinC/ISRUC cohorts under linear probing and few-shot fine-tuning. Related-work self-citations (fMRI foundation models by overlapping authors) are background, not uniqueness theorems or load-bearing premises that force the sleep results. External physiology citations motivate the prior but do not make the reported transfer scores true by construction. Under-isolation of the hierarchy prior versus scale/architecture (a validity concern) is not circularity under the stated criteria.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 1 invented entities

The central empirical claim rests on a physiological grouping prior, standard contrastive/predictive SSL machinery, and several loss/architecture knobs. No new physical particle or force is postulated; the main invented structure is the hierarchical CNS–ANS training topology and the specific positive-pair construction. Free parameters are ordinary SSL hyperparameters whose values are only partly reported.

free parameters (5)
  • lambda (L_inter weight in L_c)
    Controls strength of CNS–ANS coupling in Eq. 8; value not numerically reported but load-bearing for the hierarchy claim.
  • alpha, beta (weights on L_c and L_p)
    Overall objective Eq. 12 balances contrastive vs long-horizon prediction; schedule is two-stage but numeric weights unspecified.
  • InfoNCE temperature tau
    Standard contrastive temperature in Eq. 3; affects alignment sharpness; not reported.
  • mask set Omega / mask ratio and long-window length L
    Long-horizon MAE-style objective depends on chosen context (example 1.5h) and which epoch tokens are masked; exact settings underspecified.
  • architecture width/depth and projection head psi
    56M-parameter design choices and pooling define the representation space compared against differently sized baselines.
axioms (5)
  • domain assumption PSG modalities naturally partition into CNS (EEG, EOG, EMG) and ANS (ECG, respiration) with shared within-system factors and stage-dependent cross-system coupling.
    Introduced in Abstract/§1/§3.1 as physiological prior; supported by cited physiology but treated as fixed design, not learned topology.
  • ad hoc to paper Mean of other same-subsystem modality embeddings is a valid positive for intra-system InfoNCE (Eq. 4).
    Defines what 'intra-system consistency' means operationally; alternative positives (augmentation, temporal neighbors) are not compared.
  • domain assumption Symmetric InfoNCE between pooled CNS and ANS summaries captures brain–body synchronization useful for downstream sleep tasks (Eqs. 6–7).
    Core inter-system objective; assumes linear pooling + contrastive alignment is sufficient for coupling.
  • domain assumption Latent-space L1 prediction of masked epoch embeddings over multi-hour windows captures macro sleep dynamics better than waveform reconstruction alone (Eqs. 9–11).
    JEPA-style assumption imported from vision/SSL literature and motivated in §2.4/§3.4.
  • domain assumption Subject-specific normalization and fixed resampling (100 Hz high-frequency, 10 Hz respiratory) preserve task-relevant information across centers.
    §4.1 preprocessing assumptions underlying all transfer claims.
invented entities (1)
  • Omni-Sleep hierarchical CNS–ANS pretraining topology (intra-system + inter-system + latent long-horizon objectives) no independent evidence
    purpose: Structure multimodal PSG representation learning around physiological subsystems rather than flat fusion.
    The paper’s main methodological object; evidence is internal ablations/baselines, not an independent physiological measurement of the learned factors.

pith-pipeline@v1.1.0-grok45 · 14408 in / 3739 out tokens · 36876 ms · 2026-07-13T07:01:59.669501+00:00 · methodology

0 comments
read the original abstract

Sleep physiology arises from the coordinated dynamics of the central nervous system (CNS) and autonomic nervous system (ANS), as reflected by multimodal polysomnography signals including EEG, EOG, EMG, ECG, and respiration. However, existing sleep foundation models often fuse heterogeneous biosignals in a topology-agnostic manner, overlooking their physiological organization. We introduce Omni-Sleep, a sleep foundation model that uses the CNS/ANS partition as a physiological prior for topology-constrained representation learning. Omni-Sleep learns structured representations through three objectives: intra-system consistency, which captures shared subsystem-level factors within neural and cardio-respiratory signals; inter-system synchronization, which aligns subsystem trajectories to model brain--body dynamics; and latent-space masked temporal modeling, which captures long-horizon sleep dynamics. Pre-trained on over 100,000 hours of multi-center multimodal PSG data, Omni-Sleep is evaluated on sleep staging and multi-disease classification. Across datasets and modality-ablation settings, Omni-Sleep outperforms strong foundation-model baselines, showing improved label efficiency, cross-dataset generalization, and robustness to missing modalities. These results highlight the value of physiological hierarchy for generalizable sleep representation learning. Code is available at https://github.com/AutoBrain-sleep/OmniSleep.

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

34 extracted references · 4 linked inside Pith

  1. [1]

    Brain and brain-heart granger causality during wakefulness and sleep

    Helmi Abdalbari, Mohammad Durrani, Shivam Pancholi, Nikhil Patel, Slawomir J Nasuto, and Nicoletta Nicolaou. Brain and brain-heart granger causality during wakefulness and sleep. Frontiers in Neuroscience, 16:927111, 2022

    2022

  2. [2]

    Self-supervised learning from images with a joint-embedding predictive architecture

    Mahmoud Assran, Quentin Duval, Ishan Misra, Piotr Bojanowski, Pascal Vincent, Michael Rabbat, Yann LeCun, and Nicolas Ballas. Self-supervised learning from images with a joint-embedding predictive architecture. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15619–15629, 2023

    2023

  3. [3]

    Network physiology reveals relations between network topology and physiological function

    Amir Bashan, Ronny P Bartsch, Jan W Kantelhardt, Shlomo Havlin, and Plamen Ch Ivanov. Network physiology reveals relations between network topology and physiological function. Nature communications, 3(1):702, 2012

    2012

  4. [4]

    Racial/ethnic differences in sleep disturbances: the multi-ethnic study of atherosclerosis (mesa).Sleep, 38(6):877–888, 2015

    Xiaoli Chen, Rui Wang, Phyllis Zee, Pamela L Lutsey, Sogol Javaheri, Carmela Alcántara, Chandra L Jackson, Michelle A Williams, and Susan Redline. Racial/ethnic differences in sleep disturbances: the multi-ethnic study of atherosclerosis (mesa).Sleep, 38(6):877–888, 2015

    2015

  5. [5]

    Maeeg: Masked auto-encoder for eeg representation learning.arXiv preprint arXiv:2211.02625, 2022

    Hsiang-Yun Sherry Chien, Hanlin Goh, Christopher M Sandino, and Joseph Y Cheng. Maeeg: Masked auto-encoder for eeg representation learning.arXiv preprint arXiv:2211.02625, 2022

    Pith/arXiv arXiv 2022

  6. [6]

    Dynamic coupling between the central and autonomic nervous systems during sleep: a review

    Massimiliano de Zambotti, John Trinder, Alessandro Silvani, Ian M Colrain, and Fiona C Baker. Dynamic coupling between the central and autonomic nervous systems during sleep: a review. Neuroscience & Biobehavioral Reviews, 90:84–103, 2018. 9 Omni-Sleep: Hierarchical Contrastive Learning of CNS–ANS Dynamics

    2018

  7. [7]

    A novel sleep staging network based on data adaptation and multimodal fusion.Frontiers in human neuroscience, 15:727139, 2021

    Lijuan Duan, Mengying Li, Changming Wang, Yuanhua Qiao, Zeyu Wang, Sha Sha, and Mingai Li. A novel sleep staging network based on data adaptation and multimodal fusion.Frontiers in human neuroscience, 15:727139, 2021

    2021

  8. [8]

    Wickramaratne, Patricia Kovatch, Mayte Suarez-Farinas, Neomi A Shah, Ankit Parekh, and Girish N Nadkarni

    Benjamin Fox, Joy Jiang, Sajila D. Wickramaratne, Patricia Kovatch, Mayte Suarez-Farinas, Neomi A Shah, Ankit Parekh, and Girish N Nadkarni. A foundational transformer leveraging full night, multichannel sleep study data accurately classifies sleep stages.Sleep, 2025

    2025

  9. [9]

    An approach for determining the reliability of manual and digital scoring of sleep stages.Sleep, 46(11):zsad248, 2023

    Bethany Gerardy, Samuel T Kuna, Allan Pack, Clete A Kushida, James K Walsh, Bethany Staley, Grace W Pien, and Magdy Younes. An approach for determining the reliability of manual and digital scoring of sleep stages.Sleep, 46(11):zsad248, 2023

    2023

  10. [10]

    You snooze, you win: the physionet/computing in cardiology challenge 2018

    MohammadMGhassemi,BenjaminEMoody,Li-weiHLehman,ChristopherSong,QiaoLi,Haoqi Sun, M Brandon Westover, and Gari D Clifford. You snooze, you win: the physionet/computing in cardiology challenge 2018. In2018 Computing in Cardiology Conference (CinC), volume 45, pages 1–4. IEEE, 2018

    2018

  11. [11]

    Investigation of automated sleep staging from cardiorespiratory signals regarding clinical appli- cability and robustness.Biomedical Signal Processing and Control, 71:103047, 2022

    Miriam Goldammer, Sebastian Zaunseder, Moritz D Brandt, Hagen Malberg, and Felix Gräßer. Investigation of automated sleep staging from cardiorespiratory signals regarding clinical appli- cability and robustness.Biomedical Signal Processing and Control, 71:103047, 2022

    2022

  12. [12]

    A unified time-frequency foundation model for sleep decoding.Nature Communications, 17, 2026

    Weixuan Huang, Yan Wang, Hanrong Cheng, Wei Xu, Tingyue Li, Xiuwen Wu, Hui Xu, Pan Liao, Zaixu Cui, Qihong Zou, and Jiahong Gao. A unified time-frequency foundation model for sleep decoding.Nature Communications, 17, 2026

    2026

  13. [13]

    Large brain model for learning generic representations with tremendous eeg data in bci

    Wei-Bang Jiang, Liming Zhao, and Bao-Liang Lu. Large brain model for learning generic representations with tremendous eeg data in bci. InInternational Conference on Learning Representations, volume 2024, pages 16405–16426, 2024

    2024

  14. [14]

    Isruc-sleep: A com- prehensive public dataset for sleep researchers.Computer methods and programs in biomedicine, 124:180–192, 2016

    Sirvan Khalighi, Teresa Sousa, José Moutinho Santos, and Urbano Nunes. Isruc-sleep: A com- prehensive public dataset for sleep researchers.Computer methods and programs in biomedicine, 124:180–192, 2016

    2016

  15. [15]

    Core-sleep: A multimodal fusion framework for time series robust to imperfect modalities.IEEE Transactions on Neural Systems and Rehabilitation Engineering, 32:840–849, 2024

    Konstantinos Kontras, Christos Chatzichristos, Huy Phan, Johan Suykens, and Maarten De Vos. Core-sleep: A multimodal fusion framework for time series robust to imperfect modalities.IEEE Transactions on Neural Systems and Rehabilitation Engineering, 32:840–849, 2024

    2024

  16. [16]

    A path towards autonomous machine intelligence version 0.9

    Yann LeCun. A path towards autonomous machine intelligence version 0.9. 2, 2022-06-27. Open Review, 62(1):1–62, 2022

    2022

  17. [17]

    Automatic sleep stage classification using deep learning: signals, data representation, and neural networks

    Peng Liu, Wei Qian, Hua Zhang, Yabin Zhu, Qi Hong, Qiang Li, and Yudong Yao. Automatic sleep stage classification using deep learning: signals, data representation, and neural networks. Artificial Intelligence Review, 57(11):301, 2024

    2024

  18. [18]

    Dynamic brain-heart interaction in sleep characterized by variational phase-amplitude coupling framework.Communications Biology, 8(1):1235, 2025

    Juntong Lyu, Chien-Hung Yeh, Wenbin Shi, and Thomas Penzel. Dynamic brain-heart interaction in sleep characterized by variational phase-amplitude coupling framework.Communications Biology, 8(1):1235, 2025

    2025

  19. [19]

    A novel deep learning model based on transformer and cross modality attention for classification of sleep stages

    Sahar Hassanzadeh Mostafaei, Jafar Tanha, and Amir Sharafkhaneh. A novel deep learning model based on transformer and cross modality attention for classification of sleep stages. Journal of Biomedical Informatics, 157:104689, 2024

    2024

  20. [20]

    Xsleepnet: Multi-view sequential model for automatic sleep staging.IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(9):5903–5915, 2021

    Huy Phan, Oliver Y Chén, Minh C Tran, Philipp Koch, Alfred Mertins, and Maarten De Vos. Xsleepnet: Multi-view sequential model for automatic sleep staging.IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(9):5903–5915, 2021. 10 Omni-Sleep: Hierarchical Contrastive Learning of CNS–ANS Dynamics

    2021

  21. [21]

    Sleeptransformer: Automatic sleep staging with interpretability and uncertainty quantification

    Huy Phan, Kaare Mikkelsen, Oliver Y Chén, Philipp Koch, Alfred Mertins, and Maarten De Vos. Sleeptransformer: Automatic sleep staging with interpretability and uncertainty quantification. IEEE Transactions on Biomedical Engineering, 69(8):2456–2467, 2022

    2022

  22. [22]

    The sleep heart health study: design, rationale, and methods.Sleep, 20(12):1077–1085, 1997

    Stuart F Quan, Barbara V Howard, Conrad Iber, James P Kiley, F Javier Nieto, George T O’Connor, David M Rapoport, Susan Redline, John Robbins, Jonathan M Samet, et al. The sleep heart health study: design, rationale, and methods.Sleep, 20(12):1077–1085, 1997

    1997

  23. [23]

    Hogeon Seo, Seunghyeok Back, Seongju Lee, Deokhwan Park, Tae Kim, and Kyoobin Lee. Intra-and inter-epoch temporal context network (iitnet) using sub-epoch features for automatic sleep scoring on raw single-channel eeg.Biomedical signal processing and control, 61:102037, 2020

    2020

  24. [24]

    Sleepfm: Multi-modal representation learning for sleep across ecg, eeg and respiratory signals

    Rahul Thapa, Bryan He, Magnus Ruud Kjaer, Hyatt Moore IV, Gauri Ganjoo, Emmanuel Mignot, and James Y Zou. Sleepfm: Multi-modal representation learning for sleep across ecg, eeg and respiratory signals. InAAAI 2024 Spring Symposium on Clinical Foundation Models, 2024

    2024

  25. [25]

    Brandon Westover, Poul Jennum, Andreas Brink-Kjaer, Emmanuel Mignot, James Zou, and Magnus Rønneberg Ruud

    Rahul Thapa, Magnus Ruud Kjaer, Bryan He, Ian Covert, IV HyattMoore, Umaer Hanif, Gauri Ganjoo, M. Brandon Westover, Poul Jennum, Andreas Brink-Kjaer, Emmanuel Mignot, James Zou, and Magnus Rønneberg Ruud. A multimodal sleep foundation model for disease prediction. Nature Medicine, 32:752 – 762, 2026

    2026

  26. [26]

    An electrocardiogram-based technique to assess cardiopulmonary coupling during sleep.Sleep, 28(9):1151–1161, 2005

    Robert Joseph Thomas, Joseph E Mietus, Chung-Kang Peng, and Ary L Goldberger. An electrocardiogram-based technique to assess cardiopulmonary coupling during sleep.Sleep, 28(9):1151–1161, 2005

    2005

  27. [27]

    Slim-brain: A data-and training-efficient foundation model for fmri data analysis

    Mo Wang, Junfeng Xia, Wenhao Ye, Enyu Liu, Kaining Peng, Jianfeng Feng, Quanying Liu, and Hongkai Wen. Slim-brain: A data-and training-efficient foundation model for fmri data analysis. arXiv preprint arXiv:2512.21881, 2025

    arXiv 2025

  28. [28]

    Flexibrain: Resolution-agnostic voxel-level encoding for native fmri.arXiv preprint arXiv:2606.11500, 2026

    Mo Wang, Wenhao Ye, Junfeng Xia, Minghao Xu, Hongkai Wen, and Quanying Liu. Flexibrain: Resolution-agnostic voxel-level encoding for native fmri.arXiv preprint arXiv:2606.11500, 2026

    Pith/arXiv arXiv 2026

  29. [29]

    Omni-fmri: A universal atlas-free fmri foundation model

    Mo Wang, Wenhao Ye, Junfeng Xia, Junxiang Zhang, Xuanye Pan, Minghao Xu, Haotian Deng, Hongkai Wen, and Quanying Liu. Omni-fmri: A universal atlas-free fmri foundation model. arXiv preprint arXiv:2601.23090, 2026

    arXiv 2026

  30. [30]

    Brain-dit: A universal multi-state fmri foundation model with metadata-conditioned pretraining.arXiv preprint arXiv:2604.12683, 2026

    Junfeng Xia, Wenhao Ye, Xuanye Pan, Xinke Shen, Mo Wang, and Quanying Liu. Brain-dit: A universal multi-state fmri foundation model with metadata-conditioned pretraining.arXiv preprint arXiv:2604.12683, 2026

    Pith/arXiv arXiv 2026

  31. [31]

    Brainworld: Astructural-prior-conditionedgenerativemodelforwhole-brain4dfmridynamics.arXivpreprint arXiv:2606.17742, 2026

    JunfengXia, WenhaoYe, JunxiangZhang, XuanyePan,MoWang, andQuanyingLiu. Brainworld: Astructural-prior-conditionedgenerativemodelforwhole-brain4dfmridynamics.arXivpreprint arXiv:2606.17742, 2026

    Pith/arXiv arXiv 2026

  32. [32]

    Quantifying sleep architecture dynamics and individual differences using big data and bayesian networks.PloS one, 13(4):e0194604, 2018

    Benjamin D Yetton, Elizabeth A McDevitt, Nicola Cellini, Christian Shelton, and Sara C Mednick. Quantifying sleep architecture dynamics and individual differences using big data and bayesian networks.PloS one, 13(4):e0194604, 2018

    2018

  33. [33]

    Terry Young. Rationale, design, and findings from the wisconsin sleep cohort study: toward understanding the total societal burden of sleep-disordered breathing.Sleep medicine clinics, 4(1):37–46, 2009. 11 Omni-Sleep: Hierarchical Contrastive Learning of CNS–ANS Dynamics

    2009

  34. [34]

    Research and application of deep learning-based sleep staging: Data, modeling, validation, and clinical practice.Sleep Medicine Reviews, 74:101897, 2024

    Huijun Yue, Zhuqi Chen, Wenbin Guo, Lin Sun, Yidan Dai, Yiming Wang, Wenjun Ma, Xiaomao Fan, Weiping Wen, and Wenbin Lei. Research and application of deep learning-based sleep staging: Data, modeling, validation, and clinical practice.Sleep Medicine Reviews, 74:101897, 2024. 12

    2024