SleepMaMi (SNU; macro+micro PSG sleep foundation model, ICML 2026)
- Year
- 2026
- Form factor
- Clinical
Activity counts measure attention, not model quality. Missing counts are unknown. Publication status follows the linked source; a listed venue alone does not confirm peer review.
SleepMaMi is a polysomnography sleep foundation model with a micro-encoder over 30-second multimodal epochs and a bidirectional Mamba macro-encoder over full-night embeddings trained with demographic-guided contrastive learning. It reports strong linear-probe sleep-staging results on SHHS and CFS. Code and checkpoints are linked from the ICML 2026 / arXiv 2602.07628 release.
- Signals
- ECG, SpO₂, Resp, EEG, EMG, EOG
- Form factor
- Clinical
- Year
- 2026
- Architecture
Micro-Encoder: shared-private Transformer over 30-s epochs of 6 PSG modalities (hybrid MAE + multimodal contrastive). Macro-Encoder: bidirectional Mamba (forward + backward PreNorm-Mamba blocks) over full-night micro embeddings, trained with Demographic-Guided Contrastive Learning (age/BMI/sex distance as soft targets)
- Pretraining data
20,964 PSGs / 158,028 h: SHHS1 (train split) + SHHS2, KISS (train split), KVSS, PhysioNet 2018, MESA, MrOS
- License
MIT (code repository); checkpoints: no separate licence stated
- Type
- Foundation model
- Form factor
- Clinical
- Parameters
- Not reported
- Architecture
- Micro-Encoder: shared-private Transformer over 30-s epochs of 6 PSG modalities (hybrid MAE + multimodal contrastive). Macro-Encoder: bidirectional Mamba (forward + backward PreNorm-Mamba blocks) over full-night micro embeddings, trained with Demographic-Guided Contrastive Learning (age/BMI/sex distance as soft targets)
- Venue
- ICML 2026 (PMLR 306); arXiv v3 2026-10-22
- Year
- 2026
- DOI
- Not reported
- Weights
- Open
- Licence
- MIT (code repository); checkpoints: no separate licence stated
- Caveats
- Not reported
- Datasets
- Added
- Oct 23, 2026
- Updated
- Oct 26, 2026
Abstract
While the shift toward unified foundation models has revolutionized many deep learning domains, sleep medicine remains largely restricted to task-specific models that focus on localized micro-structure features. These approaches often neglect the rich, multi-modal context of Polysomnography (PSG) and fail to capture the global macro-structure of a full night's sleep. To address this, we introduce SleepMaMi , a Sleep Foundation Model engineered to master both hour-long sleep architectures and fine-grained signal morphologies. Our framework utilizes a hierarchical dual-encoder design: a Macro-Encoder to model full-night temporal dependencies and a Micro-Encoder to capture short-term characteristics from biosignals. Macro-Encoder is trained via Demographic-Guided Contrastive Learning, which aligns overnight sleep patterns with objective subject metadata, such as age, sex and BMI to refine global representations. Micro-Encoder is optimized via a hybrid Masked Autoencoder (MAE) and multi-modal contrastive objective. Pre-trained on a massive corpus of >20,000 PSG recordings (158K hours),SleepMaMi outperforms or matches state-of-the-art existing foundation models across a diverse suite of downstream tasks, demonstrating superior generalizability and label-efficient adaptation for clinical sleep analysis.
Verbatim fromarXiv· arXiv metadata (abstracts are released under CC0)
Key results
How it works
Datasets linked in the hub; pretraining vs evaluation roles follow the paper.
- Pretraining data
- 20,964 PSGs / 158,028 h: SHHS1 (train split) + SHHS2, KISS (train split), KVSS, PhysioNet 2018, MESA, MrOS
- Architecture
- Micro-Encoder: shared-private Transformer over 30-s epochs of 6 PSG modalities (hybrid MAE + multimodal contrastive). Macro-Encoder: bidirectional Mamba (forward + backward PreNorm-Mamba blocks) over full-night micro embeddings, trained with Demographic-Guided Contrastive Learning (age/BMI/sex distance as soft targets)
- Inputs
- ECG, SpO₂, Resp, EEG, EMG, EOG
- Approach
- foundation model · sleep · PSG · Mamba · MAE · contrastive
Leaderboard rows· 2
All leaderboards| Task | Dataset | Metric | Value | Rank |
|---|---|---|---|---|
| Sleep staging | Cleveland Family Study (CFS) | Macro-F1 ↑ | 71.2 | #1 / 1 |
| Sleep staging | Sleep Heart Health Study (SHHS, NSRR) | Macro-F1 ↑ | 74.1 | #1 / 1 |
Datasets· 4
- Sleep Heart Health Study (SHHS, NSRR)ECG · SpO₂ · Resp
- MESA SleepNSRR polysomnography + wrist actigraphy (Multi-Ethnic Study of Atherosclerosis)2,237 subjects · PPG · ECG · IMU
- MrOS Sleep Study (Osteoporotic Fractures in Men, NSRR)ECG · SpO₂ · Resp
- Cleveland Family Study (CFS)NSRR family sleep-apnea PSG2,284 subjects · PPG · PSG
Papers· 1
Cite
APA
Park, K., Na, Y., Choi, Y. R., Ryu, H., Shin, H.-W., & Kim, H.-S. (2026). SleepMaMi: A Universal Sleep Foundation Model for Integrating Macro- and Micro-structures. ICML 2026. https://arxiv.org/abs/2602.07628
BibTeX
@inproceedings{park2026sleepmami,
title = {{SleepMaMi: A Universal Sleep Foundation Model for Integrating Macro- and Micro-structures}},
author = {Park, Keondo and Na, Younghoon and Choi, You Rim and Ryu, Hyunwoo and Shin, Hyun-Woo and Kim, Hyung-Sin},
year = {2026},
booktitle = {ICML 2026},
eprint = {2602.07628},
archivePrefix = {arXiv}
}Built from the hub record. Check it against the publisher before submitting.
History
Record updated
Added to the ChatPPG research hub
Connections
Open in the graph explorerPapers
SleepMaMi: A Universal Sleep Foundation Model for Integrating Macro- and Micro-structuresPaper on this modelICML 2026 · 2026
A Unified Flexible Large Polysomnography Model for Sleep Staging and Downstream Clinical Screening (LPSGM)Also uses Sleep Heart Health Study (SHHS, NSRR)medRxiv preprint · 2024
