Paper
Rethinking ECG Tokenization: A Unified Taxonomy and Clinically-Grounded ECG Foundation Model
NeurIPS 2026 Workshop GenAI4Health (poster), OpenReview 6iqTIcGBqt, published 2026-10-26
TL;DR
ECG Tokenization Taxonomy: shared terms for describing and comparing ECG tokenizers (raw waveform, fixed-duration patches, and others). ECGTokenizer: splits each lead into physiologically meaningful regions (P wave, QRS complex, T wave, and PR, ST, TP segments) instead of fixed patches, so the irregular timing of cardiac events is part of the input. ECGTFormer: a *compact causal* ECG FM with continuous-time rotary position encoding and V-JEPA self-supervised pretraining. Takes single- or multi-lead input. The causal design allows streaming inference with prefix caching.
- Form factor: other
Results on ChatPPG leaderboards
All leaderboardsNo results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.
Tasks & signals
Signals
Keywords
Datasets
Models
How it works
Cite this paper
APA
Zhylko, D., Shamout, F. E., & list, P. T. O. A. (2026). Rethinking ECG Tokenization: A Unified Taxonomy and Clinically-Grounded ECG Foundation Model. NeurIPS 2026 Workshop GenAI4Health , OpenReview 6iqTIcGBqt. https://openreview.net/forum?id=6iqTIcGBqt
BibTeX
@inproceedings{zhylko2026rethinking,
title = {{Rethinking ECG Tokenization: A Unified Taxonomy and Clinically-Grounded ECG Foundation Model}},
author = {Zhylko, Dmytro and Shamout, Farah E. and list, per the OpenReview author},
year = {2026},
booktitle = {NeurIPS 2026 Workshop GenAI4Health , OpenReview 6iqTIcGBqt},
url = {https://openreview.net/forum?id=6iqTIcGBqt}
}Built from the hub record. Check it against the publisher before submitting.
Added to the ChatPPG research hub
