Paper

Relational Window Stabilization for Repeatable Single-Window PPG Disease-Background Prediction (RWS)

Under review2026Added Oct 5, 2026

Authors: Anonymous (ICLR 2027 double-blind)

ICLR 2027 submission (under review)

TL;DR

Argues AUROC/AUPRC miss a key continuous-monitoring property: repeated windows from the same person in an unchanged state should give repeatable predictions. RWS trains with multiple windows from the same monitoring episode (relational supervision) while keeping single-window inference — no inference-time aggregation. On the MC-MED PPG benchmark (nine chronic ICD-derived disease-background tasks, patient-disjoint chronological splits) it improves AUROC, AUPRC and ICC and lowers prediction variability and diagnostic flip rates across durations, architectures and seeds; repeatability gain persists on MIMIC-III after transfer and retraining.

Form factor: finger

Results on ChatPPG leaderboards

All leaderboards

No results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.

Tasks & signals

Signals

PPG

Keywords

repeatabilityICCchronic-diseaseICD-taskscontinuous-monitoringevaluationICLR-2027-sub

How it works

Cite this paper

APA

Relational Window Stabilization for Repeatable Single-Window PPG Disease-Background Prediction (RWS). (2026). ICLR 2027 submission. https://openreview.net/forum?id=0AQtX4yuoQ

BibTeX

@misc{relationalwindow2026relational,
  title         = {{Relational Window Stabilization for Repeatable Single-Window PPG Disease-Background Prediction (RWS)}},
  year          = {2026},
  howpublished  = {ICLR 2027 submission},
  url           = {https://openreview.net/forum?id=0AQtX4yuoQ}
}

Built from the hub record. Check it against the publisher before submitting.

Oct 5, 2026Added to the ChatPPG research hub