ChatPPG.
Join free

PaperarXiv:2602.13857

sleep2vec: Unified Cross-Modal Alignment for Heterogeneous Nocturnal Biosignals

PreprintDated Feb 2026Added Oct 19, 2026

  • WY
  • ZJ
  • YW
  • DX
  • ZY
  • CZ
  • XM
  • T
  • +2

Authors: Weixuan Yuan, Zengrui Jin, Yichen Wang, Donglin Xie, Ziyi Ye, Chao Zhang, Xuesong Chen (Five Seasons Medical, Tsinghua, PKU, Fudan)

arXiv preprint

PDF (opens in a new tab) arXiv2602.13857 (opens in a new tab) Cite

TL;DR

Contrastive pretraining on 42,249 nights (30,852 subjects) from SHHS, MrOS, MESA, WSC and HSP, with APPLES as the external cohort. Nine modalities: EEG/EOG/EMG/ECG at 128 Hz; airflow, belts, SpO2, IBI and RESP at 4 Hz. Uses 30 s tokens and a modality-agnostic RoFormer (63.5M / 133.7M / 238.2M params).

Abstract

Tasks ranging from sleep staging to clinical diagnosis traditionally rely on standard polysomnography (PSG) devices, bedside monitors and wearable devices, which capture diverse nocturnal biosignals (e.g., EEG, EOG, ECG, SpO_2). However, heterogeneity across devices and frequent sensor dropout pose significant challenges for unified modelling of these multimodal signals. We present sleep2vec, a foundation model for diverse and incomplete nocturnal biosignals that learns a shared representation via cross-modal alignment. sleep2vec is contrastively pre-trained on 42,249 overnight recordings spanning nine modalities using a Demography, Age, Site & History-aware InfoNCE objective that incorporates physiological and acquisition metadata (e.g., age, gender, recording site) to dynamically weight negatives and mitigate cohort-specific shortcuts. On downstream sleep staging and clinical outcome assessment, sleep2vec consistently outperforms strong baselines and remains robust to any subset of available modalities and sensor dropout. We further characterize, to our knowledge for the first time, scaling laws for nocturnal biosignals with respect to modality diversity and model capacity. Together, these results show that unified cross-modal alignment, coupled with principled scaling, enables label-efficient, general-purpose modelling of real-world nocturnal biosignals.

Verbatim from arXiv (opens in a new tab) · arXiv metadata (abstracts are released under CC0)

Results on ChatPPG leaderboards · 1

All leaderboards
LeaderboardMethodResultRank
shhs Cohen's kappa (5class)Sleep stagingsleep2vec0.75 (acc 83.0, MF1 65.9); APPLES ext. 0.61Cohen's kappa, lower is betterUnranked

1 row is credited to this paper. Ranks apply within each leaderboard only, as reported by the source.

Tasks & signals

Tasks

Sleep staging

Signals

ECGSpO₂RespEEGEMGEOGHRV

Devices

Philips Respironics Alice 4 PSG system

Keywords

SleepPSG FmContrastiveCross Modal AlignmentScalingSHHSMESAMrOS

Datasets used

How it works

How Method works: Setup → Method → TargetSetup: Signals (EEG, EOG, EMG, ECG, nasal-airflow, respiratory-effort, SpO2, IBI). Method: Method (Contrastive learning). Target: Sleep staging.SetupSignals: EEG, EOG, EMG, ECG, nasal-airflow, respiratory-effort, SpO2, IBISignalsEEG, EOG, EMG, ECG, nasal-airflow…MethodMethodContrastive learningTargetSleep stagingSleep staging

Cite this paper

Cite this paper with the fields recorded here. Check it against the publisher before submitting.

Yuan, W., Jin, Z., Wang, Y., Xie, D., Ye, Z., Zhang, C., & Chen, X. (2026). sleep2vec: Unified Cross-Modal Alignment for Heterogeneous Nocturnal Biosignals. arXiv preprint. https://arxiv.org/abs/2602.13857
Yuan, Weixuan, Zengrui Jin, Yichen Wang, Donglin Xie, Ziyi Ye, Chao Zhang, and Xuesong Chen. 2026. "sleep2vec: Unified Cross-Modal Alignment for Heterogeneous Nocturnal Biosignals." arXiv preprint. https://arxiv.org/abs/2602.13857.
Yuan, W., Jin, Z., Wang, Y., Xie, D., Ye, Z., Zhang, C. and Chen, X. (2026) 'sleep2vec: Unified Cross-Modal Alignment for Heterogeneous Nocturnal Biosignals', arXiv preprint. Available at: https://arxiv.org/abs/2602.13857
Yuan, Weixuan, et al. "sleep2vec: Unified Cross-Modal Alignment for Heterogeneous Nocturnal Biosignals." arXiv preprint, 2026. https://arxiv.org/abs/2602.13857.
W. Yuan et al., "sleep2vec: Unified Cross-Modal Alignment for Heterogeneous Nocturnal Biosignals," arXiv preprint, 2026, [Online]. Available: https://arxiv.org/abs/2602.13857.
@misc{yuan2026sleep2vec,
  title         = {{sleep2vec: Unified Cross-Modal Alignment for Heterogeneous Nocturnal Biosignals}},
  author        = {Yuan, Weixuan and Jin, Zengrui and Wang, Yichen and Xie, Donglin and Ye, Ziyi and Zhang, Chao and Chen, Xuesong},
  year          = {2026},
  howpublished  = {arXiv preprint},
  eprint        = {2602.13857},
  archivePrefix = {arXiv}
}

Source metadata: 2026

Record updated Oct 19, 2026

Added to the ChatPPG research hub