PPG-Bench: A Unified Benchmark for PPG Foundation Models
PPG-Bench finds most PPG foundation models fail to beat simple controls across 26 tasks.
PPG-Bench is an open-source benchmark comparing five public photoplethysmography foundation models across nine datasets and 26 tasks under linear probing, random forests, frozen MLP probes, and end-to-end finetuning. Across 68 task-regime combinations, 35.3% show no meaningful advantage over a random or constant baseline, and in 83.8% no PPG-specific model beats both pulse-rate-variability features and the MOMENT time-series model. The authors conclude current PPG foundation models do not yet provide reliably transferable representations.
- Benchmarks five public PPG foundation models on nine datasets and 26 tasks.
- Tests linear probes, random forests, frozen MLPs, and finetuning.
- 35.3% of 68 combinations show no gain over a trivial baseline.
- 83.8% of cases fail to beat both PRV features and MOMENT.
From the original article
1 min read at arxiv.orgPhotoplethysmography (PPG) is increasingly recorded by wearable and clinical devices at enormous scale, and a growing number of PPG foundation models (FMs) have been developed to compute reusable representations for various downstream tasks. These models are difficult to compare: different evaluations use different datasets, task definitions, preprocessing, downstream heads, hyperparameter tuning, and baselines.
By Arvasu Kulkarni · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.