Results explorer

Name: ChatPPG wearable AI results
Creator: ChatPPG
License: https://creativecommons.org/licenses/by/4.0/

One row per reported result from wearable PPG, ECG, motion and sleep papers. Filter by task, dataset, metric, sensor and year, or show only SOTA rows. Each value links to its source: a paper table or figure, or the leaderboard row it came from.

Last updated Oct 27, 2026

433
reported results
224
with a table or figure reference
92
papers and models
134
task × dataset × metric pairs
38
benchmarks with a SOTA star

SOTA only

433 results

Reported results with task, dataset, metric, value and source
			Metric		Comparability	Sensor / site		Source	Links
ActiNet In ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning	Activity recognition	CAPTURE-24	Cohen's κ	0.860↑	nested 5-fold group (participant) CVwithin-datasetfine-tuned + HMM smoothing30 s windows4-class activity intensity	ACC · wrist	2025	Table 1 (opens in a new tab)	Leaderboard
Random forest + HMM (accelerometer package) Baseline in ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning	Activity recognition	CAPTURE-24	Cohen's κ	0.800↑	nested 5-fold group (participant) CVwithin-datasetfine-tuned + HMM smoothing30 s windows4-class activity intensity	ACC · wrist	—	Table 1 (as reported) (opens in a new tab)
ActiNet In ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning	Activity recognition	CAPTURE-24	Macro F1	0.818↑	nested 5-fold group (participant) CVwithin-datasetfine-tuned + HMM smoothing30 s windows4-class activity intensity	ACC · wrist	2025	Table 1 (opens in a new tab)	Leaderboard
Random forest + HMM (accelerometer package) Baseline in ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning	Activity recognition	CAPTURE-24	Macro F1	0.764↑	nested 5-fold group (participant) CVwithin-datasetfine-tuned + HMM smoothing30 s windows4-class activity intensity	ACC · wrist	—	Table 1 (as reported) (opens in a new tab)
CABiGRU + diffusion synthetic pretraining In Diffusion-Based Synthetic Data Pretraining for Enhancing Activity Recognition	Activity recognition	DEO (drinking / eating / other)	Balanced accuracy (%)	90.6↑	subject-wise splitwithin-datasetpretrained on diffusion-generated smartwatch IMU, fine-tuned on real data	IMU	2026	Results (opens in a new tab)
Random forest + hidden Markov model In Statistical machine learning of sleep and physical activity phenotypes from sensor data in 96,220 UK Biobank participants	Activity recognition	Free-living wrist accelerometry (132 adults)	Accuracy (%)	87↑	not stated in the abstractwithin-datasetmachine learning on wrist accelerometer featuresphysical activity and sleep behaviours	ACC	2018	Abstract (Results) (opens in a new tab)
ActiNet In ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning	Activity recognition	RealWorld	Macro F1	0.661↑	whole dataset as external testcross-dataset (trained on CAPTURE-24)fine-tuned + HMM smoothing30 s windows4-class activity intensity	ACC · wrist	2025	Table 2 (opens in a new tab)	Leaderboard
Random forest + HMM (accelerometer package) Baseline in ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning	Activity recognition	RealWorld	Macro F1	0.531↑	whole dataset as external testcross-dataset (trained on CAPTURE-24)fine-tuned + HMM smoothing30 s windows4-class activity intensity	ACC · wrist	—	Table 2 (as reported) (opens in a new tab)
TransfHAR In TransfHAR: Self-Supervised Wrist Representations for On-Demand Activity Recognition	Activity recognition	TransfHAR in-lab study (10 participants, 7 novel wrist activities)	Balanced accuracy (%)	86.7↑	per-participant few-shot: five labeled examples per classwithin-datasetself-supervised wrist IMU representation + few-shot classifier	IMU	2026	Results, in-lab study (opens in a new tab)	Code (opens in a new tab)Hugging Face (opens in a new tab)
NormWear In NormWear: Toward Foundation Model for Multivariate Wearable Sensing of Physiological Signals	Activity recognition	UCI-HAR	AUROC (%)	98.954%↑	held-out test setwithin-datasetlinear probe, full-shot	ACC, Gyro	2024	Table 2 (opens in a new tab)	Hugging Face (opens in a new tab)
TF-C Baseline in NormWear: Toward Foundation Model for Multivariate Wearable Sensing of Physiological Signals	Activity recognition	UCI-HAR	AUROC (%)	96.892%↑	held-out test setwithin-datasetlinear probe, full-shot	ACC, Gyro	2022	Table 2 (as reported) (opens in a new tab)
CLAP Baseline in NormWear: Toward Foundation Model for Multivariate Wearable Sensing of Physiological Signals	Activity recognition	UCI-HAR	AUROC (%)	96.420%↑	held-out test setwithin-datasetlinear probe, full-shot	ACC, Gyro	2023	Table 2 (as reported) (opens in a new tab)
Statistical Baseline in NormWear: Toward Foundation Model for Multivariate Wearable Sensing of Physiological Signals	Activity recognition	UCI-HAR	AUROC (%)	95.784%↑	held-out test setwithin-datasetlinear probe, full-shot	ACC, Gyro	—	Table 2 (as reported) (opens in a new tab)
Chronos Baseline in NormWear: Toward Foundation Model for Multivariate Wearable Sensing of Physiological Signals	Activity recognition	UCI-HAR	AUROC (%)	91.593%↑	held-out test setwithin-datasetlinear probe, full-shot	ACC, Gyro	2024	Table 2 (as reported) (opens in a new tab)
ActiNet In ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning	Activity recognition	WISDM	Macro F1	0.741↑	whole dataset as external testcross-dataset (trained on CAPTURE-24)fine-tuned + HMM smoothing30 s windowsactivity intensity	ACC · wrist	2025	Table 2 (opens in a new tab)
Random forest + HMM (accelerometer package) Baseline in ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning	Activity recognition	WISDM	Macro F1	0.679↑	whole dataset as external testcross-dataset (trained on CAPTURE-24)fine-tuned + HMM smoothing30 s windowsactivity intensity	ACC · wrist	—	Table 2 (as reported) (opens in a new tab)
VitalAgent VitalAgent: A Tool-Augmented Agent for Reactive and Proactive Physiological Monitoring over Wearable Health Data	AFib / arrhythmia	Icentia11k single-lead continuous raw ECG (PhysioNet)	Balanced accuracy (HAR / class-imbalance)	0.678↑	Protocol not structured	ECG	2026	Leaderboard row	Code (opens in a new tab)Leaderboard
Pulse2View Pulse2View: Same-Sensor Cross-View Pre-Training for Patient-Level Multi-Disease Classification from Wearable PPG	AFib / arrhythmia	liu-ppg-arrhythmia-6class	Area under ROC	0.9697↑	Protocol not structured	PPG	2026	Leaderboard row	Leaderboard
VitalAgent VitalAgent: A Tool-Augmented Agent for Reactive and Proactive Physiological Monitoring over Wearable Health Data	AFib / arrhythmia	Long-term ECG and wrist PPG with annotated AF episodes (AFPPGECG)	Balanced accuracy (HAR / class-imbalance)	0.672↑	subject-disjoint	PPG · wrist	2026	Leaderboard row	Code (opens in a new tab)Leaderboard
PulseLM PulseLM: A Foundation Dataset and Benchmark for PPG-Text Learning	AFib / arrhythmia	Long-term ECG and wrist PPG with annotated AF episodes (AFPPGECG)	exact_match	0.861↑	subject-disjoint	PPG, ECG, IMU · wrist	2026	Leaderboard row	Code (opens in a new tab)Leaderboard
TF-MambaNet TF-MambaNet: A Temporal and Frequency Fused Bidirectional Mamba Architecture for PPG Foundation Model	AFib / arrhythmia	MIMIC PERform	Accuracy	0.83↑	uses extra training data	PPG, ECG, RESP, ABP · finger	2026	Leaderboard row	Leaderboard
PulseLM PulseLM: A Foundation Dataset and Benchmark for PPG-Text Learning	AFib / arrhythmia	MIMIC PERform	exact_match	0.750↑	subject-disjoint	PPG, ECG, RESP, ABP · finger	2026	Leaderboard row	Code (opens in a new tab)Leaderboard
PhysioSkeptic PhysioSkeptic: Skeptical Reasoning Grounded in Signal Quality for Multimodal Rhythm Analysis	AFib / arrhythmia	MIMIC-III-Ext-PPG	Macro-averaged F1	0.886↑	Protocol not structured	PPG, ECG · finger	2026	Leaderboard row	Leaderboard
MS-ECG-FM MS-ECG-FM: Towards a More Universal Electrocardiogram Foundation Model for Health Monitoring using Multi-source Contrastive Learning	AFib / arrhythmia	ptb-xl-rhythm	Area under ROC	0.987↑	uses extra training data	ECG	2026	Leaderboard row	Leaderboard
Random Forest (George 2026) Development of an efficient deep learning model for arrhythmia detection from consumer smartwatches	AFib / arrhythmia	Pulsewatch	Area under ROC	0.9703↑	Protocol not structured	PPG, ECG, IMU · wrist	2026	Leaderboard row	Leaderboard
Random Forest, engineered features Development of an efficient deep learning model for arrhythmia detection from consumer smartwatches	AFib / arrhythmia	Pulsewatch	Macro-averaged F1	0.842↑	subject-disjoint	PPG, IMU · wrist	2026	Leaderboard row	Leaderboard
xECG BenchECG and xECG: a benchmark and baseline for ECG foundation models (xLSTM)	AFib / arrhythmia	Stanford wrist PPG AF (DeepBeat)	Area under ROC	0.883↑	uses extra training data	PPG, ECG · wrist	2026	Leaderboard row	Code (opens in a new tab)Leaderboard
xECG BenchECG and xECG: a benchmark and baseline for ECG foundation models (xLSTM)	AFib / arrhythmia	Stanford wrist PPG AF (DeepBeat)	Area under ROC	0.751↑	uses extra training data	PPG, ECG · wrist	2026	Leaderboard row	Code (opens in a new tab)Leaderboard
ECG-PPG JEPA 345M Bridging ECG and PPG: Latent-Space Prediction for Robust Physiological Analysis	AFib / arrhythmia	Stanford wrist PPG AF (DeepBeat)	Macro-averaged F1	0.834↑	uses extra training data	PPG · wrist	2026	Leaderboard row	Leaderboard
AnyPPG In AnyPPG: An ECG-Guided PPG Foundation Model Trained on Over 100,000 Hours of Recordings for Holistic Health Profiling	AFib detection	DeepBeat	AUROC	0.69↑	as defined per dataset in the paperwithin-datasetlinear probe on frozen embeddings	PPG · wrist	2025	Table 4 (opens in a new tab)	Code (opens in a new tab)
SimCLR Baseline in AnyPPG: An ECG-Guided PPG Foundation Model Trained on Over 100,000 Hours of Recordings for Holistic Health Profiling	AFib detection	DeepBeat	AUROC	0.68↑	as defined per dataset in the paperwithin-datasetlinear probe on frozen embeddings	PPG · wrist	2020	Table 4 (as reported) (opens in a new tab)
Chronos-2 Baseline in AnyPPG: An ECG-Guided PPG Foundation Model Trained on Over 100,000 Hours of Recordings for Holistic Health Profiling	AFib detection	DeepBeat	AUROC	0.66↑	as defined per dataset in the paperwithin-datasetlinear probe on frozen embeddings	PPG · wrist	2025	Table 4 (as reported) (opens in a new tab)
BYOL Baseline in AnyPPG: An ECG-Guided PPG Foundation Model Trained on Over 100,000 Hours of Recordings for Holistic Health Profiling	AFib detection	DeepBeat	AUROC	0.64↑	as defined per dataset in the paperwithin-datasetlinear probe on frozen embeddings	PPG · wrist	2020	Table 4 (as reported) (opens in a new tab)
MOMENT Baseline in AnyPPG: An ECG-Guided PPG Foundation Model Trained on Over 100,000 Hours of Recordings for Holistic Health Profiling	AFib detection	DeepBeat	AUROC	0.60↑	as defined per dataset in the paperwithin-datasetlinear probe on frozen embeddings	PPG · wrist	2024	Table 4 (as reported) (opens in a new tab)
PulsePPG Baseline in AnyPPG: An ECG-Guided PPG Foundation Model Trained on Over 100,000 Hours of Recordings for Holistic Health Profiling	AFib detection	DeepBeat	AUROC	0.60↑	as defined per dataset in the paperwithin-datasetlinear probe on frozen embeddings	PPG · wrist	2025	Table 4 (as reported) (opens in a new tab)	Code (opens in a new tab)
PaPaGEI-S Baseline in AnyPPG: An ECG-Guided PPG Foundation Model Trained on Over 100,000 Hours of Recordings for Holistic Health Profiling	AFib detection	DeepBeat	AUROC	0.57↑	as defined per dataset in the paperwithin-datasetlinear probe on frozen embeddings	PPG · wrist	2024	Table 4 (as reported) (opens in a new tab)	Code (opens in a new tab)
ECGFounder In ECGFounder: An ECG Foundation Model Built on 10+ Million Recordings	AFib detection	External single-lead devices	AUROC	0.957↑	external test setcross-datasetpretrained classifierAF, single-lead	ECG · wearable / portable ECG	2024	Results, single-lead external test (opens in a new tab)	Code (opens in a new tab)Leaderboard
ECGFounder In ECGFounder: An ECG Foundation Model Built on 10+ Million Recordings	AFib detection	PTB-XL	AUROC	0.993↑	external test set (not used in training)cross-datasetpretrained classifier, no fine-tuningatrial fibrillation, 12-lead	ECG	2024	Table 2 (opens in a new tab)	Code (opens in a new tab)
CTN Baseline in ECGFounder: An ECG Foundation Model Built on 10+ Million Recordings	AFib detection	PTB-XL	AUROC	0.953↑	external test set (not used in training)cross-datasetpretrained classifier, no fine-tuningatrial fibrillation, 12-lead	ECG	2020	Table 2 (as reported) (opens in a new tab)
ECG-SE-ResNet Baseline in ECGFounder: An ECG Foundation Model Built on 10+ Million Recordings	AFib detection	PTB-XL	AUROC	0.950↑	external test set (not used in training)cross-datasetpretrained classifier, no fine-tuningatrial fibrillation, 12-lead	ECG	2020	Table 2 (as reported) (opens in a new tab)
S12L-ECG Baseline in ECGFounder: An ECG Foundation Model Built on 10+ Million Recordings	AFib detection	PTB-XL	AUROC	0.925↑	external test set (not used in training)cross-datasetpretrained classifier, no fine-tuningatrial fibrillation, 12-lead	ECG	2020	Table 2 (as reported) (opens in a new tab)
GPT-PPG 1B In GPT-PPG: a GPT-based foundation model for photoplethysmography signals	AFib detection	Stanford AF (DeepBeat)	F1 score	0.847↑	predefined subject-wise train/val/testwithin-datasetfine-tuned	PPG · wrist	2025	Table 1 (opens in a new tab)
GPT-PPG 345M Baseline in GPT-PPG: a GPT-based foundation model for photoplethysmography signals	AFib detection	Stanford AF (DeepBeat)	F1 score	0.841↑	predefined subject-wise train/val/testwithin-datasetfine-tuned	PPG · wrist	2025	Table 1 (as reported) (opens in a new tab)
SiamQuality Baseline in GPT-PPG: a GPT-based foundation model for photoplethysmography signals	AFib detection	Stanford AF (DeepBeat)	F1 score	0.710↑	predefined subject-wise train/val/testwithin-datasetfine-tuned	PPG · wrist	2024	Table 1 (as reported) (opens in a new tab)
Shen et al. Baseline in GPT-PPG: a GPT-based foundation model for photoplethysmography signals	AFib detection	Stanford AF (DeepBeat)	F1 score	0.684↑	predefined subject-wise train/val/testwithin-datasetfine-tuned	PPG · wrist	2019	Table 1 (as reported) (opens in a new tab)
BayesBeat Baseline in GPT-PPG: a GPT-based foundation model for photoplethysmography signals	AFib detection	Stanford AF (DeepBeat)	F1 score	0.671↑	predefined subject-wise train/val/testwithin-datasetfine-tuned	PPG · wrist	2020	Table 1 (as reported) (opens in a new tab)
DeepBeat Baseline in GPT-PPG: a GPT-based foundation model for photoplethysmography signals	AFib detection	Stanford AF (DeepBeat)	F1 score	0.652↑	predefined subject-wise train/val/testwithin-datasetfine-tuned	PPG · wrist	2020	Table 1 (as reported) (opens in a new tab)
ECG-FM (Full) In ECG-FM: An Open Electrocardiogram Foundation Model	AFib detection	UHN-ECG	AUROC	0.996↑	patient-stratified ~80/10/10 train/validation/test split (no patient overlap)within-datasetpretrained ECG foundation model, fine-tunedatrial fibrillation label of the ECG interpretation task	ECG	2024	Table 3 (opens in a new tab)
PaPaGei-S In PaPaGei: Open Foundation Models for Optical Physiological Signals	Apnea-hypopnea index	MESA	MAE (events/h)	12.97↓	subject-level train/val/testwithin-datasetlinear probe on frozen embeddings	PPG	2024	Tables 3 and 4 (opens in a new tab)	Code (opens in a new tab)
PaPaGei-P Baseline in PaPaGei: Open Foundation Models for Optical Physiological Signals	Apnea-hypopnea index	MESA	MAE (events/h)	13.85↓	subject-level train/val/testwithin-datasetlinear probe on frozen embeddings	PPG	2024	Tables 3 and 4 (as reported) (opens in a new tab)

ActiNet

In ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning

0.860↑

Activity recognition · CAPTURE-24 · Cohen's κ

nested 5-fold group (participant) CVwithin-datasetfine-tuned + HMM smoothing30 s windows4-class activity intensity

Table 1

2025

Random forest + HMM (accelerometer package)

Baseline in ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning

0.800↑

Activity recognition · CAPTURE-24 · Cohen's κ

nested 5-fold group (participant) CVwithin-datasetfine-tuned + HMM smoothing30 s windows4-class activity intensity

Table 1 (as reported)

ActiNet

In ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning

0.818↑

Activity recognition · CAPTURE-24 · Macro F1

nested 5-fold group (participant) CVwithin-datasetfine-tuned + HMM smoothing30 s windows4-class activity intensity

Table 1

2025

Random forest + HMM (accelerometer package)

Baseline in ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning

0.764↑

Activity recognition · CAPTURE-24 · Macro F1

nested 5-fold group (participant) CVwithin-datasetfine-tuned + HMM smoothing30 s windows4-class activity intensity

Table 1 (as reported)

CABiGRU + diffusion synthetic pretraining

In Diffusion-Based Synthetic Data Pretraining for Enhancing Activity Recognition

90.6↑

Activity recognition · DEO (drinking / eating / other) · Balanced accuracy (%)

subject-wise splitwithin-datasetpretrained on diffusion-generated smartwatch IMU, fine-tuned on real data

Results

2026

Random forest + hidden Markov model

In Statistical machine learning of sleep and physical activity phenotypes from sensor data in 96,220 UK Biobank participants

87↑

Activity recognition · Free-living wrist accelerometry (132 adults) · Accuracy (%)

not stated in the abstractwithin-datasetmachine learning on wrist accelerometer featuresphysical activity and sleep behaviours

Abstract (Results)

2018

ActiNet

In ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning

0.661↑

Activity recognition · RealWorld · Macro F1

whole dataset as external testcross-dataset (trained on CAPTURE-24)fine-tuned + HMM smoothing30 s windows4-class activity intensity

Table 2

2025

Random forest + HMM (accelerometer package)

Baseline in ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning

0.531↑

Activity recognition · RealWorld · Macro F1

whole dataset as external testcross-dataset (trained on CAPTURE-24)fine-tuned + HMM smoothing30 s windows4-class activity intensity

Table 2 (as reported)

TransfHAR

In TransfHAR: Self-Supervised Wrist Representations for On-Demand Activity Recognition

86.7↑

Activity recognition · TransfHAR in-lab study (10 participants, 7 novel wrist activities) · Balanced accuracy (%)

per-participant few-shot: five labeled examples per classwithin-datasetself-supervised wrist IMU representation + few-shot classifier

Results, in-lab study

2026

NormWear

In NormWear: Toward Foundation Model for Multivariate Wearable Sensing of Physiological Signals

98.954%↑

Activity recognition · UCI-HAR · AUROC (%)

held-out test setwithin-datasetlinear probe, full-shot

Table 2

2024

TF-C

Baseline in NormWear: Toward Foundation Model for Multivariate Wearable Sensing of Physiological Signals

96.892%↑

Activity recognition · UCI-HAR · AUROC (%)

held-out test setwithin-datasetlinear probe, full-shot

Table 2 (as reported)

2022

CLAP

Baseline in NormWear: Toward Foundation Model for Multivariate Wearable Sensing of Physiological Signals

96.420%↑

Activity recognition · UCI-HAR · AUROC (%)

held-out test setwithin-datasetlinear probe, full-shot

Table 2 (as reported)

2023

Statistical

Baseline in NormWear: Toward Foundation Model for Multivariate Wearable Sensing of Physiological Signals

95.784%↑

Activity recognition · UCI-HAR · AUROC (%)

held-out test setwithin-datasetlinear probe, full-shot

Table 2 (as reported)

Chronos

Baseline in NormWear: Toward Foundation Model for Multivariate Wearable Sensing of Physiological Signals

91.593%↑

Activity recognition · UCI-HAR · AUROC (%)

held-out test setwithin-datasetlinear probe, full-shot

Table 2 (as reported)

2024

ActiNet

In ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning

0.741↑

Activity recognition · WISDM · Macro F1

whole dataset as external testcross-dataset (trained on CAPTURE-24)fine-tuned + HMM smoothing30 s windowsactivity intensity

Table 2

2025

Random forest + HMM (accelerometer package)

Baseline in ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning

0.679↑

Activity recognition · WISDM · Macro F1

whole dataset as external testcross-dataset (trained on CAPTURE-24)fine-tuned + HMM smoothing30 s windowsactivity intensity

Table 2 (as reported)

VitalAgent

VitalAgent: A Tool-Augmented Agent for Reactive and Proactive Physiological Monitoring over Wearable Health Data

0.678↑

AFib / arrhythmia · Icentia11k single-lead continuous raw ECG (PhysioNet) · Balanced accuracy (HAR / class-imbalance)

Protocol not structured

Leaderboard row

2026

Pulse2View

Pulse2View: Same-Sensor Cross-View Pre-Training for Patient-Level Multi-Disease Classification from Wearable PPG

0.9697↑

AFib / arrhythmia · liu-ppg-arrhythmia-6class · Area under ROC

Protocol not structured

Leaderboard row

2026

VitalAgent

VitalAgent: A Tool-Augmented Agent for Reactive and Proactive Physiological Monitoring over Wearable Health Data

0.672↑

AFib / arrhythmia · Long-term ECG and wrist PPG with annotated AF episodes (AFPPGECG) · Balanced accuracy (HAR / class-imbalance)

subject-disjoint

Leaderboard row

2026

PulseLM

PulseLM: A Foundation Dataset and Benchmark for PPG-Text Learning

0.861↑

AFib / arrhythmia · Long-term ECG and wrist PPG with annotated AF episodes (AFPPGECG) · exact_match

subject-disjoint

Leaderboard row

2026

TF-MambaNet

TF-MambaNet: A Temporal and Frequency Fused Bidirectional Mamba Architecture for PPG Foundation Model

0.83↑

AFib / arrhythmia · MIMIC PERform · Accuracy

uses extra training data

Leaderboard row

2026

PulseLM

PulseLM: A Foundation Dataset and Benchmark for PPG-Text Learning

0.750↑

AFib / arrhythmia · MIMIC PERform · exact_match

subject-disjoint

Leaderboard row

2026

PhysioSkeptic

PhysioSkeptic: Skeptical Reasoning Grounded in Signal Quality for Multimodal Rhythm Analysis

0.886↑

AFib / arrhythmia · MIMIC-III-Ext-PPG · Macro-averaged F1

Protocol not structured

Leaderboard row

2026

MS-ECG-FM

MS-ECG-FM: Towards a More Universal Electrocardiogram Foundation Model for Health Monitoring using Multi-source Contrastive Learning

0.987↑

AFib / arrhythmia · ptb-xl-rhythm · Area under ROC

uses extra training data

Leaderboard row

2026

Random Forest (George 2026)

Development of an efficient deep learning model for arrhythmia detection from consumer smartwatches

0.9703↑

AFib / arrhythmia · Pulsewatch · Area under ROC

Protocol not structured

Leaderboard row

2026

Random Forest, engineered features

Development of an efficient deep learning model for arrhythmia detection from consumer smartwatches

0.842↑

AFib / arrhythmia · Pulsewatch · Macro-averaged F1

subject-disjoint

Leaderboard row

2026

xECG

BenchECG and xECG: a benchmark and baseline for ECG foundation models (xLSTM)

0.883↑

AFib / arrhythmia · Stanford wrist PPG AF (DeepBeat) · Area under ROC

uses extra training data

Leaderboard row

2026

xECG

BenchECG and xECG: a benchmark and baseline for ECG foundation models (xLSTM)

0.751↑

AFib / arrhythmia · Stanford wrist PPG AF (DeepBeat) · Area under ROC

uses extra training data

Leaderboard row

2026

ECG-PPG JEPA 345M

Bridging ECG and PPG: Latent-Space Prediction for Robust Physiological Analysis

0.834↑

AFib / arrhythmia · Stanford wrist PPG AF (DeepBeat) · Macro-averaged F1

uses extra training data

Leaderboard row

2026

AnyPPG

In AnyPPG: An ECG-Guided PPG Foundation Model Trained on Over 100,000 Hours of Recordings for Holistic Health Profiling

0.69↑

AFib detection · DeepBeat · AUROC

as defined per dataset in the paperwithin-datasetlinear probe on frozen embeddings

Table 4

2025

SimCLR

Baseline in AnyPPG: An ECG-Guided PPG Foundation Model Trained on Over 100,000 Hours of Recordings for Holistic Health Profiling

0.68↑

AFib detection · DeepBeat · AUROC

as defined per dataset in the paperwithin-datasetlinear probe on frozen embeddings

Table 4 (as reported)

2020

Chronos-2

Baseline in AnyPPG: An ECG-Guided PPG Foundation Model Trained on Over 100,000 Hours of Recordings for Holistic Health Profiling

0.66↑

AFib detection · DeepBeat · AUROC

as defined per dataset in the paperwithin-datasetlinear probe on frozen embeddings

Table 4 (as reported)

2025

BYOL

Baseline in AnyPPG: An ECG-Guided PPG Foundation Model Trained on Over 100,000 Hours of Recordings for Holistic Health Profiling

0.64↑

AFib detection · DeepBeat · AUROC

as defined per dataset in the paperwithin-datasetlinear probe on frozen embeddings

Table 4 (as reported)

2020

MOMENT

Baseline in AnyPPG: An ECG-Guided PPG Foundation Model Trained on Over 100,000 Hours of Recordings for Holistic Health Profiling

0.60↑

AFib detection · DeepBeat · AUROC

as defined per dataset in the paperwithin-datasetlinear probe on frozen embeddings

Table 4 (as reported)

2024

PulsePPG

Baseline in AnyPPG: An ECG-Guided PPG Foundation Model Trained on Over 100,000 Hours of Recordings for Holistic Health Profiling

0.60↑

AFib detection · DeepBeat · AUROC

as defined per dataset in the paperwithin-datasetlinear probe on frozen embeddings

Table 4 (as reported)

2025

PaPaGEI-S

Baseline in AnyPPG: An ECG-Guided PPG Foundation Model Trained on Over 100,000 Hours of Recordings for Holistic Health Profiling

0.57↑

AFib detection · DeepBeat · AUROC

as defined per dataset in the paperwithin-datasetlinear probe on frozen embeddings

Table 4 (as reported)

2024

ECGFounder

In ECGFounder: An ECG Foundation Model Built on 10+ Million Recordings

0.957↑

AFib detection · External single-lead devices · AUROC

external test setcross-datasetpretrained classifierAF, single-lead

Results, single-lead external test

2024

ECGFounder

In ECGFounder: An ECG Foundation Model Built on 10+ Million Recordings

0.993↑

AFib detection · PTB-XL · AUROC

external test set (not used in training)cross-datasetpretrained classifier, no fine-tuningatrial fibrillation, 12-lead

Table 2

2024

CTN

Baseline in ECGFounder: An ECG Foundation Model Built on 10+ Million Recordings

0.953↑

AFib detection · PTB-XL · AUROC

external test set (not used in training)cross-datasetpretrained classifier, no fine-tuningatrial fibrillation, 12-lead

Table 2 (as reported)

2020

ECG-SE-ResNet

Baseline in ECGFounder: An ECG Foundation Model Built on 10+ Million Recordings

0.950↑

AFib detection · PTB-XL · AUROC

external test set (not used in training)cross-datasetpretrained classifier, no fine-tuningatrial fibrillation, 12-lead

Table 2 (as reported)

2020

S12L-ECG

Baseline in ECGFounder: An ECG Foundation Model Built on 10+ Million Recordings

0.925↑

AFib detection · PTB-XL · AUROC

external test set (not used in training)cross-datasetpretrained classifier, no fine-tuningatrial fibrillation, 12-lead

Table 2 (as reported)

2020

GPT-PPG 1B

In GPT-PPG: a GPT-based foundation model for photoplethysmography signals

0.847↑

AFib detection · Stanford AF (DeepBeat) · F1 score

predefined subject-wise train/val/testwithin-datasetfine-tuned

Table 1

2025

GPT-PPG 345M

Baseline in GPT-PPG: a GPT-based foundation model for photoplethysmography signals

0.841↑

AFib detection · Stanford AF (DeepBeat) · F1 score

predefined subject-wise train/val/testwithin-datasetfine-tuned

Table 1 (as reported)

2025

SiamQuality

Baseline in GPT-PPG: a GPT-based foundation model for photoplethysmography signals

0.710↑

AFib detection · Stanford AF (DeepBeat) · F1 score

predefined subject-wise train/val/testwithin-datasetfine-tuned

Table 1 (as reported)

2024

Shen et al.

Baseline in GPT-PPG: a GPT-based foundation model for photoplethysmography signals

0.684↑

AFib detection · Stanford AF (DeepBeat) · F1 score

predefined subject-wise train/val/testwithin-datasetfine-tuned

Table 1 (as reported)

2019

BayesBeat

Baseline in GPT-PPG: a GPT-based foundation model for photoplethysmography signals

0.671↑

AFib detection · Stanford AF (DeepBeat) · F1 score

predefined subject-wise train/val/testwithin-datasetfine-tuned

Table 1 (as reported)

2020

DeepBeat

Baseline in GPT-PPG: a GPT-based foundation model for photoplethysmography signals

0.652↑

AFib detection · Stanford AF (DeepBeat) · F1 score

predefined subject-wise train/val/testwithin-datasetfine-tuned

Table 1 (as reported)

2020

ECG-FM (Full)

In ECG-FM: An Open Electrocardiogram Foundation Model

0.996↑

AFib detection · UHN-ECG · AUROC

patient-stratified ~80/10/10 train/validation/test split (no patient overlap)within-datasetpretrained ECG foundation model, fine-tunedatrial fibrillation label of the ECG interpretation task

Table 3

2024

PaPaGei-S

In PaPaGei: Open Foundation Models for Optical Physiological Signals

12.97↓

Apnea-hypopnea index · MESA · MAE (events/h)

subject-level train/val/testwithin-datasetlinear probe on frozen embeddings

Tables 3 and 4

2024

PaPaGei-P

Baseline in PaPaGei: Open Foundation Models for Optical Physiological Signals

13.85↓

Apnea-hypopnea index · MESA · MAE (events/h)

subject-level train/val/testwithin-datasetlinear probe on frozen embeddings

Tables 3 and 4 (as reported)

2024

Showing 1–50 of 433 results

Load 50 more results

Recent changes

Oct 27, 2026
New entry: A cross-subject stress recognition method…, 0.9158
Stress detection (acute / prolonged / affect): WESAD: accuracy
Oct 27, 2026
New entry: SPAFT, 5.066
Heart rate estimation: PPG-DaLiA: MAE (BPM)
Oct 27, 2026
New entry: SPAFT, 8.345
Cuffless blood pressure estimation: uci-bp: MAE (mmHg)
Oct 25, 2026
Leaderboard added with 1 result
Cuffless blood pressure estimation: neo-bp: MAE (mmHg)
Oct 23, 2026
New entry: PC-AVCT wearable-state model, 7.697
Cuffless blood pressure estimation: aurora-bp: MAE (mmHg)
Oct 23, 2026
New entry: PPG-FusionNet, 7.46
Cuffless blood pressure estimation: pulsedb-calbased: MAE (mmHg)

From the hub's dated sync history: new results as they are added, and SOTA stars as they are awarded or move.

How SOTA is decided

marks the best value for one task, dataset and metric, using the metric's direction (higher or lower is better).
Results are compared only under the same evaluation protocol: split type, within- or cross-dataset testing, filtering, inputs and label set. Different protocols stay separate and are never ranked together.
Only results with a table, figure or section reference compete, and a star needs at least two competing methods. Ties go to the earlier publication.
Stars are recomputed whenever the research hub syncs, and every change is logged.

Downloads of every result (JSON and CSV) are free for members. Sign in to download.

Results as reported by the authors; not independently reproduced. Leaderboard rows without a structured protocol are listed but don't receive stars. See also the leaderboards.

Cite or reuse this

CC BY 4.0ChatPPG compiled this data. You can copy, adapt and republish it, including in articles, slides, papers and AI tools, as long as you credit ChatPPG and link to this page. Please also cite the original studies for any number you use.

Attribution

Source: ChatPPG, Wearable AI results explorer, https://chatppg.com/research/results (CC BY 4.0)

HTML link

Source: <a href="https://chatppg.com/research/results">ChatPPG, Wearable AI results explorer</a> (<a href="https://creativecommons.org/licenses/by/4.0/" rel="license">CC BY 4.0</a>)