Model

PPG-Distill

From “PPG-Distill: Efficient Photoplethysmography Signals Analysis via Foundation Model Distillation”
PPGNeurIPS 2025 Workshop on Learning from TimOpen codeWeights: not released
Venue listedCatalog paper date: 2025Added to atlas: Oct 25, 2026Source

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.

PPG-Distill is a knowledge-distillation framework that compresses PPG foundation teachers such as GPT-PPG-19m and PaPaGei into roughly 1M-parameter students for edge devices. It adds morphology and rhythm distillation on top of prediction and feature KD so heart-rate and atrial-fibrillation performance stay close to the teacher at a fraction of the cost.

Venue
NeurIPS 2025 TS4H Workshop (Spotlight)
Signals
PPG
Authors
Ni, Kataria, Tang, Yang, Hu, Jin (Emory)
Teachers
GPT-PPG-19m, PaPaGei → students GPT-PPG-1m / MLP
Code
github.com/xiaoyuli0575/PPG-Distill
Year
2025

DaLiA HR MAE 8.34 bpm; StanfordAF F1 0.77 (+21.8% relative)

Up to 7× faster inference and 19× lower memory vs teacher

At a glance
Type
FM adaptation
Modalities
PPG
Parameters
1M
Architecture
Prediction + feature + patch morphology (InfoNCE) + rhythm (relational) KD; students GPT-PPG-1m / MLP; teachers GPT-PPG-19m / PaPaGei
Venue
NeurIPS 2025 Workshop on Learning from Time Series for Health (TS4H Spotlight)
Year
2025
DOI
Not reported
Weights
Not released
Licence
Not reported
Caveats
Not reported
Added
Oct 25, 2026
Updated
Oct 25, 2026
PPG-Distill: ChatPPG research summary card

How it works

How PPG-Distill works: Input signals → Model → Downstream tasksInput signals: PPG (Photoplethysmogram). Datasets: PPG-DaLiA, Stanford wrist PPG AF (DeepBeat). Model: PPG-Distill, 1M parameters.Input signalsPPG: PhotoplethysmogramPPGPhotoplethysmogramDatasets: PPG-DaLiA, Stanford wrist PPG AF (DeepBeat)DatasetsPPG-DaLiA, Stanford wrist PPG AF (DeepBeat)ModelPPG-Distill1M parameters

Datasets linked in the hub; pretraining vs evaluation roles follow the paper.

Architecture
Prediction + feature + patch morphology (InfoNCE) + rhythm (relational) KD; students GPT-PPG-1m / MLP; teachers GPT-PPG-19m / PaPaGei
Inputs
PPG
Approach
PPG · distillation · knowledge distillation · edge · emory · neurips 2025

Devices· 2

Full photo credits and licences are on each device page. An icon means we have no openly licensed or manufacturer-cleared photo of that exact device yet.

Datasets· 2

Papers· 1

Cite

APA

Ni, J., Kataria, S., Tang, S., Yang, C., Hu, X., & Jin, W. (2025). PPG-Distill: Efficient Photoplethysmography Signals Analysis via Foundation Model Distillation. NeurIPS 2025 Workshop on Learning from Time Series for Health. https://arxiv.org/abs/2509.19215

BibTeX

@inproceedings{ni2025ppgdistill,
  title         = {{PPG-Distill: Efficient Photoplethysmography Signals Analysis via Foundation Model Distillation}},
  author        = {Ni, Juntong and Kataria, Saurabh and Tang, Shengpu and Yang, Carl and Hu, Xiao and Jin, Wei},
  year          = {2025},
  booktitle     = {NeurIPS 2025 Workshop on Learning from Time Series for Health},
  eprint        = {2509.19215},
  archivePrefix = {arXiv}
}

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

History

  1. Added to the ChatPPG research hub