Weighted Temporal Decay Loss for Learning Wearable PPG Data with Sparse Clinical Labels

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chung, Yunsung, Chun, Keum San, Gwak, Migyeong, Feng, Han, Liu, Yingshuo, Lim, Chanho, Nathan, Viswam, Marrouche, Nassir, Desai, Sharanya Arcot
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908807905411072
author Chung, Yunsung
Chun, Keum San
Gwak, Migyeong
Feng, Han
Liu, Yingshuo
Lim, Chanho
Nathan, Viswam
Marrouche, Nassir
Desai, Sharanya Arcot
author_facet Chung, Yunsung
Chun, Keum San
Gwak, Migyeong
Feng, Han
Liu, Yingshuo
Lim, Chanho
Nathan, Viswam
Marrouche, Nassir
Desai, Sharanya Arcot
contents Advances in wearable computing and AI have increased interest in leveraging PPG for health monitoring over the past decade. One of the biggest challenges in developing health algorithms based on such biosignals is the sparsity of clinical labels, which makes biosignals temporally distant from lab draws less reliable for supervision. To address this problem, we introduce a simple training strategy that learns a biomarker-specific decay of sample weight over the time gap between a segment and its ground truth label and uses this weight in the loss with a regularizer to prevent trivial solutions. On smartwatch PPG from 450 participants across 10 biomarkers, the approach improves over baselines. In the subject-wise setting, the proposed approach averages 0.715 AUPRC, compared to 0.674 for a fine-tuned self-supervised baseline and 0.626 for a feature-based Random Forest. A comparison of four decay families shows that a simple linear decay function is most robust on average. Beyond accuracy, the learned decay rates summarize how quickly each biomarker's PPG evidence becomes stale, providing an interpretable view of temporal sensitivity.
format Preprint
id arxiv_https___arxiv_org_abs_2602_02917
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Weighted Temporal Decay Loss for Learning Wearable PPG Data with Sparse Clinical Labels
Chung, Yunsung
Chun, Keum San
Gwak, Migyeong
Feng, Han
Liu, Yingshuo
Lim, Chanho
Nathan, Viswam
Marrouche, Nassir
Desai, Sharanya Arcot
Machine Learning
Advances in wearable computing and AI have increased interest in leveraging PPG for health monitoring over the past decade. One of the biggest challenges in developing health algorithms based on such biosignals is the sparsity of clinical labels, which makes biosignals temporally distant from lab draws less reliable for supervision. To address this problem, we introduce a simple training strategy that learns a biomarker-specific decay of sample weight over the time gap between a segment and its ground truth label and uses this weight in the loss with a regularizer to prevent trivial solutions. On smartwatch PPG from 450 participants across 10 biomarkers, the approach improves over baselines. In the subject-wise setting, the proposed approach averages 0.715 AUPRC, compared to 0.674 for a fine-tuned self-supervised baseline and 0.626 for a feature-based Random Forest. A comparison of four decay families shows that a simple linear decay function is most robust on average. Beyond accuracy, the learned decay rates summarize how quickly each biomarker's PPG evidence becomes stale, providing an interpretable view of temporal sensitivity.
title Weighted Temporal Decay Loss for Learning Wearable PPG Data with Sparse Clinical Labels
topic Machine Learning
url https://arxiv.org/abs/2602.02917