{"generated_at":"2026-10-16T14:01:47-05:00","count":191,"source":"https://chatppg.com/research","tasks":[{"task":"glucose-estimation","task_name":"Glucose estimation","total_rows":58,"boards":[{"id":"glucose-estimation__ppg-bgl-kermani__clarke_a_pct","dataset":"ppg-bgl-kermani","dataset_label":"PPG Signals for Blood Glucose Measurement (Kermani / Mendeley)","metric":"clarke_a_pct","metric_label":"Clarke error grid zone A (%)","higher_is_better":true,"rows":[{"rank":1,"method":"PPG features + partial least squares (Islam et al. 2021 method) [reproduction]","method_short":"PLS (Islam)","entry_slug":null,"entry_kind":null,"value":86.57,"value_text":"86.57","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":2,"method":"MFCC features + XGBoost (Prabha et al. 2022 method) [reproduction]","method_short":"MFCC + XGBoost (Prabha)","entry_slug":null,"entry_kind":null,"value":83.58,"value_text":"83.58","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":3,"method":"Mean-of-training-set baseline [reproduction]","method_short":"Mean baseline","entry_slug":null,"entry_kind":null,"value":77.61,"value_text":"77.61","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":4,"method":"ResNet34, 1 s segments (Zeynali et al. 2025)","method_short":"ResNet34-1s","entry_slug":null,"entry_kind":null,"value":76.6,"value_text":"76.6","protocol":"Original paper: trained on VitalDB (random 70/15/15 split), external test on the MUST (Kermani) dataset with 1 s peak-centred segments resampled to 100 Hz (Tables 6-7).","open_weights":null},{"rank":5,"method":"PPG features + XGBoost (Gupta et al. 2021 method) [reproduction]","method_short":"XGBoost (Gupta)","entry_slug":null,"entry_kind":null,"value":75.6,"value_text":"75.6","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":6,"method":"ResNet34 on raw 1 s PPG (Zeynali et al. 2025 method) [reproduction]","method_short":"ResNet34 (Zeynali)","entry_slug":null,"entry_kind":null,"value":72.17,"value_text":"72.17","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":7,"method":"VGG16, 1 s segments (Zeynali et al. 2025)","method_short":"VGG16-1s","entry_slug":null,"entry_kind":null,"value":71.3,"value_text":"71.3","protocol":"Original paper: trained on VitalDB (random 70/15/15 split), external test on the MUST (Kermani) dataset with 1 s peak-centred segments resampled to 100 Hz (Tables 6-7).","open_weights":null},{"rank":8,"method":"CNN-LSTM-Attention, 1 s segments (Zeynali et al. 2025)","method_short":"CNN-LSTM-Att-1s","entry_slug":null,"entry_kind":null,"value":69,"value_text":"69","protocol":"Original paper: trained on VitalDB (random 70/15/15 split), external test on the MUST (Kermani) dataset with 1 s peak-centred segments resampled to 100 Hz (Tables 6-7).","open_weights":null},{"rank":9,"method":"1D-CNN on raw PPG (Alghlayini et al. 2023 method) [reproduction]","method_short":"1D-CNN (Alghlayini)","entry_slug":null,"entry_kind":null,"value":62.69,"value_text":"62.69","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null}]},{"id":"glucose-estimation__ppg-bgl-kermani__mae_mgdl","dataset":"ppg-bgl-kermani","dataset_label":"PPG Signals for Blood Glucose Measurement (Kermani / Mendeley)","metric":"mae_mgdl","metric_label":"Glucose MAE (mg/dL)","higher_is_better":false,"rows":[{"rank":1,"method":"PPG features + partial least squares (Islam et al. 2021 method) [reproduction]","method_short":"PLS (Islam)","entry_slug":null,"entry_kind":null,"value":13.37,"value_text":"13.37","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":2,"method":"MFCC features + XGBoost (Prabha et al. 2022 method) [reproduction]","method_short":"MFCC + XGBoost (Prabha)","entry_slug":null,"entry_kind":null,"value":14.23,"value_text":"14.23","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":3,"method":"ResNet34, 1 s segments (Zeynali et al. 2025)","method_short":"ResNet34-1s","entry_slug":null,"entry_kind":null,"value":14.8,"value_text":"14.8","protocol":"Original paper: trained on VitalDB (random 70/15/15 split), external test on the MUST (Kermani) dataset with 1 s peak-centred segments resampled to 100 Hz (Tables 6-7).","open_weights":null},{"rank":4,"method":"Mean-of-training-set baseline [reproduction]","method_short":"Mean baseline","entry_slug":null,"entry_kind":null,"value":16,"value_text":"16","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":5,"method":"ResNet34 on raw 1 s PPG (Zeynali et al. 2025 method) [reproduction]","method_short":"ResNet34 (Zeynali)","entry_slug":null,"entry_kind":null,"value":16.33,"value_text":"16.33","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":6,"method":"PPG features + XGBoost (Gupta et al. 2021 method) [reproduction]","method_short":"XGBoost (Gupta)","entry_slug":null,"entry_kind":null,"value":16.69,"value_text":"16.69","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":7,"method":"VGG16, 1 s segments (Zeynali et al. 2025)","method_short":"VGG16-1s","entry_slug":null,"entry_kind":null,"value":17,"value_text":"17","protocol":"Original paper: trained on VitalDB (random 70/15/15 split), external test on the MUST (Kermani) dataset with 1 s peak-centred segments resampled to 100 Hz (Tables 6-7).","open_weights":null},{"rank":8,"method":"CNN-LSTM-Attention, 1 s segments (Zeynali et al. 2025)","method_short":"CNN-LSTM-Att-1s","entry_slug":null,"entry_kind":null,"value":17.1,"value_text":"17.1","protocol":"Original paper: trained on VitalDB (random 70/15/15 split), external test on the MUST (Kermani) dataset with 1 s peak-centred segments resampled to 100 Hz (Tables 6-7).","open_weights":null},{"rank":9,"method":"1D-CNN on raw PPG (Alghlayini et al. 2023 method) [reproduction]","method_short":"1D-CNN (Alghlayini)","entry_slug":null,"entry_kind":null,"value":18.21,"value_text":"18.21","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null}]},{"id":"glucose-estimation__ppg-bgl-kermani__mard_pct","dataset":"ppg-bgl-kermani","dataset_label":"PPG Signals for Blood Glucose Measurement (Kermani / Mendeley)","metric":"mard_pct","metric_label":"MARD (%)","higher_is_better":false,"rows":[{"rank":1,"method":"PPG features + partial least squares (Islam et al. 2021 method) [reproduction]","method_short":"PLS (Islam)","entry_slug":null,"entry_kind":null,"value":10.67,"value_text":"10.67","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":2,"method":"MFCC features + XGBoost (Prabha et al. 2022 method) [reproduction]","method_short":"MFCC + XGBoost (Prabha)","entry_slug":null,"entry_kind":null,"value":11.69,"value_text":"11.69","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":3,"method":"ResNet34, 1 s segments (Zeynali et al. 2025)","method_short":"ResNet34-1s","entry_slug":null,"entry_kind":null,"value":12.8,"value_text":"12.8","protocol":"Original paper: trained on VitalDB (random 70/15/15 split), external test on the MUST (Kermani) dataset with 1 s peak-centred segments resampled to 100 Hz (Tables 6-7).","open_weights":null},{"rank":4,"method":"CNN-LSTM-Attention, 1 s segments (Zeynali et al. 2025)","method_short":"CNN-LSTM-Att-1s","entry_slug":null,"entry_kind":null,"value":13,"value_text":"13","protocol":"Original paper: trained on VitalDB (random 70/15/15 split), external test on the MUST (Kermani) dataset with 1 s peak-centred segments resampled to 100 Hz (Tables 6-7).","open_weights":null},{"rank":5,"method":"Mean-of-training-set baseline [reproduction]","method_short":"Mean baseline","entry_slug":null,"entry_kind":null,"value":14.13,"value_text":"14.13","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":6,"method":"PPG features + XGBoost (Gupta et al. 2021 method) [reproduction]","method_short":"XGBoost (Gupta)","entry_slug":null,"entry_kind":null,"value":14.46,"value_text":"14.46","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":7,"method":"ResNet34 on raw 1 s PPG (Zeynali et al. 2025 method) [reproduction]","method_short":"ResNet34 (Zeynali)","entry_slug":null,"entry_kind":null,"value":14.99,"value_text":"14.99","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null},{"rank":8,"method":"VGG16, 1 s segments (Zeynali et al. 2025)","method_short":"VGG16-1s","entry_slug":null,"entry_kind":null,"value":15,"value_text":"15","protocol":"Original paper: trained on VitalDB (random 70/15/15 split), external test on the MUST (Kermani) dataset with 1 s peak-centred segments resampled to 100 Hz (Tables 6-7).","open_weights":null},{"rank":9,"method":"1D-CNN on raw PPG (Alghlayini et al. 2023 method) [reproduction]","method_short":"1D-CNN (Alghlayini)","entry_slug":null,"entry_kind":null,"value":16.56,"value_text":"16.56","protocol":"Trained on all VitalDB data, tested on the 67 MUST (Kermani) recordings (Supraja et al. 2026 out-of-distribution test). MUST glucose spans only 88-183 mg/dL, so mean prediction scores well; R² was below zero for every model.","open_weights":null}]},{"id":"glucose-estimation__vitaldb__clarke_a_pct","dataset":"vitaldb","dataset_label":"VitalDB","metric":"clarke_a_pct","metric_label":"Clarke error grid zone A (%)","higher_is_better":true,"rows":[{"rank":1,"method":"ResNet34 on raw 1 s PPG (Zeynali et al. 2025 method)","method_short":"ResNet34 (Zeynali)","entry_slug":null,"entry_kind":null,"value":55.62,"value_text":"55.62","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":2,"method":"MFCC features + XGBoost (Prabha et al. 2022 method)","method_short":"MFCC + XGBoost (Prabha)","entry_slug":null,"entry_kind":null,"value":55.1,"value_text":"55.1","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":3,"method":"1D-CNN on raw PPG (Alghlayini et al. 2023 method)","method_short":"1D-CNN (Alghlayini)","entry_slug":null,"entry_kind":null,"value":54.19,"value_text":"54.19","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":4,"method":"PPG features + XGBoost (Gupta et al. 2021 method)","method_short":"XGBoost (Gupta)","entry_slug":null,"entry_kind":null,"value":53.88,"value_text":"53.88","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":5,"method":"PPG features + partial least squares (Islam et al. 2021 method)","method_short":"PLS (Islam)","entry_slug":null,"entry_kind":null,"value":53.19,"value_text":"53.19","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":6,"method":"Mean-of-training-set baseline","method_short":"Mean baseline","entry_slug":null,"entry_kind":null,"value":52.27,"value_text":"52.27","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null}]},{"id":"glucose-estimation__vitaldb__mae_mgdl","dataset":"vitaldb","dataset_label":"VitalDB","metric":"mae_mgdl","metric_label":"Glucose MAE (mg/dL)","higher_is_better":false,"rows":[{"rank":1,"method":"MFCC features + XGBoost (Prabha et al. 2022 method)","method_short":"MFCC + XGBoost (Prabha)","entry_slug":null,"entry_kind":null,"value":25.42,"value_text":"25.42","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":2,"method":"PPG features + XGBoost (Gupta et al. 2021 method)","method_short":"XGBoost (Gupta)","entry_slug":null,"entry_kind":null,"value":25.77,"value_text":"25.77","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":3,"method":"ResNet34 on raw 1 s PPG (Zeynali et al. 2025 method)","method_short":"ResNet34 (Zeynali)","entry_slug":null,"entry_kind":null,"value":25.93,"value_text":"25.93","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":4,"method":"1D-CNN on raw PPG (Alghlayini et al. 2023 method)","method_short":"1D-CNN (Alghlayini)","entry_slug":null,"entry_kind":null,"value":26.05,"value_text":"26.05","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":5,"method":"PPG features + partial least squares (Islam et al. 2021 method)","method_short":"PLS (Islam)","entry_slug":null,"entry_kind":null,"value":26.14,"value_text":"26.14","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":6,"method":"Mean-of-training-set baseline","method_short":"Mean baseline","entry_slug":null,"entry_kind":null,"value":26.39,"value_text":"26.39","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null}]},{"id":"glucose-estimation__vitaldb__mard_pct","dataset":"vitaldb","dataset_label":"VitalDB","metric":"mard_pct","metric_label":"MARD (%)","higher_is_better":false,"rows":[{"rank":1,"method":"MFCC features + XGBoost (Prabha et al. 2022 method)","method_short":"MFCC + XGBoost (Prabha)","entry_slug":null,"entry_kind":null,"value":21.56,"value_text":"21.56","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":2,"method":"1D-CNN on raw PPG (Alghlayini et al. 2023 method)","method_short":"1D-CNN (Alghlayini)","entry_slug":null,"entry_kind":null,"value":21.8,"value_text":"21.8","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":3,"method":"PPG features + XGBoost (Gupta et al. 2021 method)","method_short":"XGBoost (Gupta)","entry_slug":null,"entry_kind":null,"value":21.96,"value_text":"21.96","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":4,"method":"PPG features + partial least squares (Islam et al. 2021 method)","method_short":"PLS (Islam)","entry_slug":null,"entry_kind":null,"value":22.28,"value_text":"22.28","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":5,"method":"Mean-of-training-set baseline","method_short":"Mean baseline","entry_slug":null,"entry_kind":null,"value":22.52,"value_text":"22.52","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null},{"rank":6,"method":"ResNet34 on raw 1 s PPG (Zeynali et al. 2025 method)","method_short":"ResNet34 (Zeynali)","entry_slug":null,"entry_kind":null,"value":22.55,"value_text":"22.55","protocol":"Reproduced by Supraja et al. 2026 on VitalDB intraoperative finger PPG vs lab glucose; leave-some-participants-out (test = 1/5 of participants, non-overlapping), mean of 5 folds. A predict-the-training-mean baseline scores within ~1 mg/dL of every model.","open_weights":null}]},{"id":"glucose-estimation__big-ideas-glycemic__mard_pct","dataset":"big-ideas-glycemic","dataset_label":"BIG IDEAs Lab Glycemic Variability and Wearable Device Data","metric":"mard_pct","metric_label":"MARD (%)","higher_is_better":false,"rows":[{"rank":1,"method":"XGBoost regression (Karunarathna & Liang 2025)","method_short":"XGBoost (Karunarathna)","entry_slug":null,"entry_kind":null,"value":7.1,"value_text":"7.1","protocol":"236 wearable + circadian + demographic features, no food logs; random 80/20 split of 5-min epochs (same participants in train and test), mean of 20 random seeds. Not subject-independent, so not comparable with LOPOCV rows.","open_weights":null},{"rank":2,"method":"Random forest regression (Karunarathna & Liang 2025)","method_short":"Random forest (Karunarathna)","entry_slug":null,"entry_kind":null,"value":9.7,"value_text":"9.7","protocol":"236 wearable + circadian + demographic features, no food logs; random 80/20 split of 5-min epochs (same participants in train and test), mean of 20 random seeds. Not subject-independent, so not comparable with LOPOCV rows.","open_weights":null},{"rank":3,"method":"Personalized gradient boosting (Bent et al. 2021)","method_short":"Bent 2021 personalized","entry_slug":null,"entry_kind":null,"value":13.26,"value_text":"13.26","protocol":"Personalized gradient-boosted model: trained on the first half of each participant's own data, tested on the second half. MARD here is the paper's MAPE against Dexcom G6.","open_weights":null},{"rank":4,"method":"Population gradient boosting, LOPOCV (Bent et al. 2021)","method_short":"Bent 2021 population","entry_slug":null,"entry_kind":null,"value":14.33,"value_text":"14.33","protocol":"Gradient-boosted model on Empatica E4 features + food logs, 5-min CGM targets; leave-one-person-out CV across 16 participants (mean over participants). MARD here is the paper's MAPE against Dexcom G6.","open_weights":null}]},{"id":"glucose-estimation__big-ideas-glycemic__rmse_mgdl","dataset":"big-ideas-glycemic","dataset_label":"BIG IDEAs Lab Glycemic Variability and Wearable Device Data","metric":"rmse_mgdl","metric_label":"Glucose RMSE (mg/dL)","higher_is_better":false,"rows":[{"rank":1,"method":"XGBoost regression (Karunarathna & Liang 2025)","method_short":"XGBoost (Karunarathna)","entry_slug":null,"entry_kind":null,"value":11.9,"value_text":"11.9","protocol":"236 wearable + circadian + demographic features, no food logs; random 80/20 split of 5-min epochs (same participants in train and test), mean of 20 random seeds. Not subject-independent, so not comparable with LOPOCV rows.","open_weights":null},{"rank":2,"method":"Random forest regression (Karunarathna & Liang 2025)","method_short":"Random forest (Karunarathna)","entry_slug":null,"entry_kind":null,"value":15.6,"value_text":"15.6","protocol":"236 wearable + circadian + demographic features, no food logs; random 80/20 split of 5-min epochs (same participants in train and test), mean of 20 random seeds. Not subject-independent, so not comparable with LOPOCV rows.","open_weights":null},{"rank":3,"method":"Personalized gradient boosting (Bent et al. 2021)","method_short":"Bent 2021 personalized","entry_slug":null,"entry_kind":null,"value":21.1,"value_text":"21.1","protocol":"Personalized gradient-boosted model: trained on the first half of each participant's own data, tested on the second half. MARD here is the paper's MAPE against Dexcom G6.","open_weights":null},{"rank":4,"method":"Population gradient boosting, LOPOCV (Bent et al. 2021)","method_short":"Bent 2021 population","entry_slug":null,"entry_kind":null,"value":21.22,"value_text":"21.22","protocol":"Gradient-boosted model on Empatica E4 features + food logs, 5-min CGM targets; leave-one-person-out CV across 16 participants (mean over participants). MARD here is the paper's MAPE against Dexcom G6.","open_weights":null}]},{"id":"glucose-estimation__ppg-bgl-kermani__rmse_mgdl","dataset":"ppg-bgl-kermani","dataset_label":"PPG Signals for Blood Glucose Measurement (Kermani / Mendeley)","metric":"rmse_mgdl","metric_label":"Glucose RMSE (mg/dL)","higher_is_better":false,"rows":[{"rank":1,"method":"ResNet34, 1 s segments (Zeynali et al. 2025)","method_short":"ResNet34-1s","entry_slug":null,"entry_kind":null,"value":19.7,"value_text":"19.7","protocol":"Original paper: trained on VitalDB (random 70/15/15 split), external test on the MUST (Kermani) dataset with 1 s peak-centred segments resampled to 100 Hz (Tables 6-7).","open_weights":null},{"rank":2,"method":"VGG16, 1 s segments (Zeynali et al. 2025)","method_short":"VGG16-1s","entry_slug":null,"entry_kind":null,"value":21,"value_text":"21","protocol":"Original paper: trained on VitalDB (random 70/15/15 split), external test on the MUST (Kermani) dataset with 1 s peak-centred segments resampled to 100 Hz (Tables 6-7).","open_weights":null},{"rank":3,"method":"CNN-LSTM-Attention, 1 s segments (Zeynali et al. 2025)","method_short":"CNN-LSTM-Att-1s","entry_slug":null,"entry_kind":null,"value":24.4,"value_text":"24.4","protocol":"Original paper: trained on VitalDB (random 70/15/15 split), external test on the MUST (Kermani) dataset with 1 s peak-centred segments resampled to 100 Hz (Tables 6-7).","open_weights":null}]},{"id":"glucose-estimation__big-ideas-glycemic__clarke_a_pct","dataset":"big-ideas-glycemic","dataset_label":"BIG IDEAs Lab Glycemic Variability and Wearable Device Data","metric":"clarke_a_pct","metric_label":"Clarke error grid zone A (%)","higher_is_better":true,"rows":[{"rank":1,"method":"XGBoost regression (Karunarathna & Liang 2025)","method_short":"XGBoost (Karunarathna)","entry_slug":null,"entry_kind":null,"value":94.2,"value_text":"94.2","protocol":"236 wearable + circadian + demographic features, no food logs; random 80/20 split of 5-min epochs (same participants in train and test), Clarke zones from one of the train-test splits (Table 3). Not subject-independent, so not comparable with LOPOCV rows.","open_weights":null},{"rank":2,"method":"Random forest regression (Karunarathna & Liang 2025)","method_short":"Random forest (Karunarathna)","entry_slug":null,"entry_kind":null,"value":89.1,"value_text":"89.1","protocol":"236 wearable + circadian + demographic features, no food logs; random 80/20 split of 5-min epochs (same participants in train and test), Clarke zones from one of the train-test splits (Table 3). Not subject-independent, so not comparable with LOPOCV rows.","open_weights":null}]}]},{"task":"har","task_name":"Human activity recognition","total_rows":57,"boards":[{"id":"har__pamap2__macro_f1_har","dataset":"pamap2","dataset_label":"PAMAP2","metric":"macro_f1_har","metric_label":"Macro-F1 on HAR activities","higher_is_better":true,"rows":[{"rank":1,"method":"ontoar-har-ontology-guided-agentic-rag-har-2026 (OntoAR-HAR, Gemini 3.0 Flash)","method_short":"OntoAR-HAR","entry_slug":"ontoar-har-ontology-guided-agentic-rag-har-2026","entry_kind":"paper","value":87.3,"value_text":"87.3 ± 1.1 (%)","protocol":"Mean of ten samples; 2026-10-14","open_weights":null},{"rank":2,"method":"chear-confusion-hierarchy-evidence-aware-llm-har-2026 (CHEAR, Gemini 3 Flash)","method_short":"CHEAR","entry_slug":"chear-confusion-hierarchy-evidence-aware-llm-har-2026","entry_kind":"paper","value":83.43,"value_text":"83.43 (%)","protocol":"vs ZARA 76.9 in the same table; not ranked against LOSO MorphCL/HARNet; 2026-10-14","open_weights":null},{"rank":3,"method":"Self-supervised ResNet pre-trained on UK Biobank, all layers fine-tuned","method_short":"SSL ResNet, UK Biobank (Yuan)","entry_slug":null,"entry_kind":null,"value":78.9,"value_text":"78.9","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Shown in % to match this board.","open_weights":null},{"rank":4,"method":"zara-training-free-motion-reasoning-llm-agents-acl-2026 (ZARA, Gemini-2.0-Flash)","method_short":"ZARA","entry_slug":"zara-training-free-motion-reasoning-llm-agents-acl-2026","entry_kind":"paper","value":76.9,"value_text":"76.9 (%)","protocol":"Held-out-subject class-balanced split, 12 classes — different protocol from the LOSO MorphCL/HARNet row; not ranked against it (protocol isolation). MorphCL LOSO row unchanged","open_weights":null},{"rank":5,"method":"Random forest on hand-crafted features","method_short":"Random forest (Yuan)","entry_slug":null,"entry_kind":null,"value":75.3,"value_text":"75.3","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Shown in % to match this board.","open_weights":null},{"rank":6,"method":"morphcl-morphological-contrastive-imu-har-2026 (MTL+MorphCL ResNet18, CAPTURE-24 3.7K h)","method_short":"MorphCL","entry_slug":"morphcl-morphological-contrastive-imu-har-2026","entry_kind":"paper","value":75.15,"value_text":"75.15 LP / 75.80 FT (%)","protocol":"LOSO, right wrist 50 Hz 2.56 s, NULL kept; comparator Yuan et al. UKB ResNet18 (harnet-ukb) 77.06 LP / 84.00 FT on same protocol (paper Table 5).","open_weights":null},{"rank":7,"method":"ResNet-V2 trained from scratch","method_short":"ResNet from scratch (Yuan)","entry_slug":null,"entry_kind":null,"value":60.5,"value_text":"60.5","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Shown in % to match this board.","open_weights":null}]},{"id":"har__opportunity__macro_f1_har","dataset":"opportunity","dataset_label":"opportunity","metric":"macro_f1_har","metric_label":"Macro-F1 on HAR activities","higher_is_better":true,"rows":[{"rank":1,"method":"zara-training-free-motion-reasoning-llm-agents-acl-2026 (ZARA, Gemini-2.0-Flash, training-free LLM agents)","method_short":"ZARA","entry_slug":"zara-training-free-motion-reasoning-llm-agents-acl-2026","entry_kind":"paper","value":92.5,"value_text":"92.5 (%)","protocol":"ZARA protocol: knowledge base from seen subjects, class-balanced held-out-subject test, 4 classes (Table 1).","open_weights":null},{"rank":2,"method":"ontoar-har-ontology-guided-agentic-rag-har-2026 (OntoAR-HAR, Gemini 3.0 Flash)","method_short":"OntoAR-HAR","entry_slug":"ontoar-har-ontology-guided-agentic-rag-har-2026","entry_kind":"paper","value":91.5,"value_text":"91.5 ± 1.6 (%)","protocol":"Mean of ten samples; 2026-10-14","open_weights":null},{"rank":3,"method":"chear-confusion-hierarchy-evidence-aware-llm-har-2026 (CHEAR, Gemini 3 Flash)","method_short":"CHEAR","entry_slug":"chear-confusion-hierarchy-evidence-aware-llm-har-2026","entry_kind":"paper","value":85.13,"value_text":"85.13 (%)","protocol":"ZARA 92.5 in the same table; 2026-10-14","open_weights":null},{"rank":4,"method":"Self-supervised ResNet pre-trained on UK Biobank, all layers fine-tuned","method_short":"SSL ResNet, UK Biobank (Yuan)","entry_slug":null,"entry_kind":null,"value":59.5,"value_text":"59.5","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Shown in % to match this board.","open_weights":null},{"rank":5,"method":"Random forest on hand-crafted features","method_short":"Random forest (Yuan)","entry_slug":null,"entry_kind":null,"value":41.6,"value_text":"41.6","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Shown in % to match this board.","open_weights":null},{"rank":6,"method":"ResNet-V2 trained from scratch","method_short":"ResNet from scratch (Yuan)","entry_slug":null,"entry_kind":null,"value":38.3,"value_text":"38.3","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Shown in % to match this board.","open_weights":null}]},{"id":"har__capture-24__macro_f1_har","dataset":"capture-24","dataset_label":"capture-24","metric":"macro_f1_har","metric_label":"Macro-F1 on HAR activities","higher_is_better":true,"rows":[{"rank":1,"method":"actinet-ssl-harnet-hmm-activity-intensity-sci-rep-2026 (ActiNet = harnet-ukb fine-tuned + HMM)","method_short":"ActiNet","entry_slug":"actinet-ssl-harnet-hmm-activity-intensity-sci-rep-2026","entry_kind":"paper","value":0.818,"value_text":"0.818","protocol":"Sleep/sedentary/light/MVPA (Walmsley), 30-s @30 Hz wrist ACC, 151 pts, nested 5-fold group CV; κ 0.860; RF+HMM 0.764.","open_weights":null},{"rank":2,"method":"Self-supervised ResNet pre-trained on UK Biobank, all layers fine-tuned","method_short":"SSL ResNet, UK Biobank (Yuan)","entry_slug":null,"entry_kind":null,"value":0.726,"value_text":"0.726","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Capture-24 4-class labels.","open_weights":null},{"rank":3,"method":"ResNet-V2 trained from scratch","method_short":"ResNet from scratch (Yuan)","entry_slug":null,"entry_kind":null,"value":0.708,"value_text":"0.708","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Capture-24 4-class labels.","open_weights":null},{"rank":4,"method":"Random forest on hand-crafted features","method_short":"Random forest (Yuan)","entry_slug":null,"entry_kind":null,"value":0.694,"value_text":"0.694","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Capture-24 4-class labels.","open_weights":null},{"rank":5,"method":"doomen-grip-open-source-wrist-acc-activity-intensity-chronic-pain-fdgth-2026 (GRIP CNN, CC BY 4.0 weights)","method_short":"An open-source deep learning model for classifyi","entry_slug":"doomen-grip-open-source-wrist-acc-activity-intensity-chronic-pain-fdgth-2026","entry_kind":"paper","value":0.661,"value_text":"0.661","protocol":"Sedentary/light/MVPA (no sleep), 10-s windows, Capture-24 downsampled 100→12.5 Hz, trained on CP+HP lab + Capture-24 (\"All on Capture\"), repeated 10-fold subject-level CV; acc 67.2 %.","open_weights":null}]},{"id":"har__capture-24__cohen_kappa","dataset":"capture-24","dataset_label":"capture-24","metric":"cohen_kappa","metric_label":"Cohen’s κ (e.g. sleep staging)","higher_is_better":true,"rows":[{"rank":1,"method":"HARNet / OxWearables SSL","method_short":"HARNet / OxWearables SSL","entry_slug":"actinet-ssl-harnet-hmm-activity-intensity-sci-rep-2026","entry_kind":"paper","value":0.86,"value_text":"0.86","protocol":"same 4-class intensity protocol as above; RF+HMM baseline 0.800","open_weights":true},{"rank":2,"method":"Self-supervised ResNet pre-trained on UK Biobank, all layers fine-tuned","method_short":"SSL ResNet, UK Biobank (Yuan)","entry_slug":null,"entry_kind":null,"value":0.737,"value_text":"0.737","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Cohen's kappa, Capture-24 4-class labels.","open_weights":null},{"rank":3,"method":"ResNet-V2 trained from scratch","method_short":"ResNet from scratch (Yuan)","entry_slug":null,"entry_kind":null,"value":0.703,"value_text":"0.703","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Cohen's kappa, Capture-24 4-class labels.","open_weights":null},{"rank":4,"method":"Random forest on hand-crafted features","method_short":"Random forest (Yuan)","entry_slug":null,"entry_kind":null,"value":0.683,"value_text":"0.683","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Cohen's kappa, Capture-24 4-class labels.","open_weights":null}]},{"id":"har__mhealth__f1_har_unspecified","dataset":"mhealth","dataset_label":"mhealth","metric":"f1_har_unspecified","metric_label":"F1 on HAR activities (averaging not stated)","higher_is_better":true,"rows":[{"rank":1,"method":"Multi-channel autoencoder, multi-task (MCAE)","method_short":"MCAE (Azadi)","entry_slug":null,"entry_kind":null,"value":99.58,"value_text":"99.58","protocol":"Held-out test subjects (UCI-HAR uses the official 21/9 subject split); F1 in %, averaging (macro or weighted) not stated for the headline table (Table 7).","open_weights":null},{"rank":2,"method":"Single-task supervised baseline","method_short":"Single-task baseline (Azadi)","entry_slug":null,"entry_kind":null,"value":99.23,"value_text":"99.23","protocol":"Held-out test subjects (UCI-HAR uses the official 21/9 subject split); F1 in %, averaging (macro or weighted) not stated for the headline table (Table 7).","open_weights":null},{"rank":3,"method":"Classical autoencoder, multi-task (CAE)","method_short":"CAE (Azadi)","entry_slug":null,"entry_kind":null,"value":97.33,"value_text":"97.33","protocol":"Held-out test subjects (UCI-HAR uses the official 21/9 subject split); F1 in %, averaging (macro or weighted) not stated for the headline table (Table 7).","open_weights":null},{"rank":4,"method":"Random-forest strategy (Yang)","method_short":"Random forest (Yang)","entry_slug":null,"entry_kind":null,"value":94.49,"value_text":"94.49","protocol":"Leave-one-subject-out; random-forest-based strategy; the paper reports an average F1 and accuracy in %.","open_weights":null}]},{"id":"har__mhealth__macro_f1_har","dataset":"mhealth","dataset_label":"mhealth","metric":"macro_f1_har","metric_label":"Macro-F1 on HAR activities","higher_is_better":true,"rows":[{"rank":1,"method":"chear-confusion-hierarchy-evidence-aware-llm-har-2026 (CHEAR, Gemini 3 Flash Preview, training-free LLM RAG)","method_short":"CHEAR","entry_slug":"chear-confusion-hierarchy-evidence-aware-llm-har-2026","entry_kind":"paper","value":88.62,"value_text":"88.62 (%)","protocol":"Cross-subject; prior-method numbers in the paper are copied from ZARA (not re-run) → same-family as the ZARA row (86.1, Gemini-2.0-Flash) but different LLM; 2026-10-14","open_weights":null},{"rank":2,"method":"zara-training-free-motion-reasoning-llm-agents-acl-2026 (ZARA, Gemini-2.0-Flash)","method_short":"ZARA","entry_slug":"zara-training-free-motion-reasoning-llm-agents-acl-2026","entry_kind":"paper","value":86.1,"value_text":"86.1 (%)","protocol":"same ZARA protocol, 12 classes.","open_weights":null},{"rank":3,"method":"ontoar-har-ontology-guided-agentic-rag-har-2026 (OntoAR-HAR, Gemini 3.0 Flash, agentic RAG + personalized memory)","method_short":"OntoAR-HAR","entry_slug":"ontoar-har-ontology-guided-agentic-rag-har-2026","entry_kind":"paper","value":84.6,"value_text":"84.6 ± 2.6 (%)","protocol":"Mean of ten random samples (Table II); ZARA row copied in-table; 2026-10-14","open_weights":null},{"rank":4,"method":"inrtlwrist (InrtlWrist)","method_short":"InrtlWrist","entry_slug":"inrtlwrist","entry_kind":"model","value":68.6,"value_text":"68.6 probe / 72.4 FT (%)","protocol":"IMU-Bench subject-disjoint 2.56 s, 12 classes, wrist channels only — different protocol from ZARA-family rows; not ranked against them.","open_weights":null}]},{"id":"har__pamap2__f1_har_unspecified","dataset":"pamap2","dataset_label":"PAMAP2","metric":"f1_har_unspecified","metric_label":"F1 on HAR activities (averaging not stated)","higher_is_better":true,"rows":[{"rank":1,"method":"Multi-channel autoencoder, multi-task (MCAE)","method_short":"MCAE (Azadi)","entry_slug":null,"entry_kind":null,"value":94.88,"value_text":"94.88","protocol":"Held-out test subjects (UCI-HAR uses the official 21/9 subject split); F1 in %, averaging (macro or weighted) not stated for the headline table (Table 7).","open_weights":null},{"rank":2,"method":"Single-task supervised baseline","method_short":"Single-task baseline (Azadi)","entry_slug":null,"entry_kind":null,"value":92.66,"value_text":"92.66","protocol":"Held-out test subjects (UCI-HAR uses the official 21/9 subject split); F1 in %, averaging (macro or weighted) not stated for the headline table (Table 7).","open_weights":null},{"rank":3,"method":"Random-forest strategy (Yang)","method_short":"Random forest (Yang)","entry_slug":null,"entry_kind":null,"value":91.64,"value_text":"91.64","protocol":"Leave-one-subject-out; random-forest-based strategy; the paper reports an average F1 and accuracy in %.","open_weights":null},{"rank":4,"method":"Classical autoencoder, multi-task (CAE)","method_short":"CAE (Azadi)","entry_slug":null,"entry_kind":null,"value":91,"value_text":"91.0","protocol":"Held-out test subjects (UCI-HAR uses the official 21/9 subject split); F1 in %, averaging (macro or weighted) not stated for the headline table (Table 7).","open_weights":null}]},{"id":"har__realworld-har__macro_f1_har","dataset":"realworld-har","dataset_label":"RealWorld HAR","metric":"macro_f1_har","metric_label":"Macro-F1 on HAR activities","higher_is_better":true,"rows":[{"rank":1,"method":"Self-supervised ResNet pre-trained on UK Biobank, all layers fine-tuned","method_short":"SSL ResNet, UK Biobank (Yuan)","entry_slug":null,"entry_kind":null,"value":0.792,"value_text":"0.792","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2).","open_weights":null},{"rank":2,"method":"Random forest on hand-crafted features","method_short":"Random forest (Yuan)","entry_slug":null,"entry_kind":null,"value":0.731,"value_text":"0.731","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2).","open_weights":null},{"rank":3,"method":"ResNet-V2 trained from scratch","method_short":"ResNet from scratch (Yuan)","entry_slug":null,"entry_kind":null,"value":0.705,"value_text":"0.705","protocol":"Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2).","open_weights":null},{"rank":4,"method":"actinet-ssl-harnet-hmm-activity-intensity-sci-rep-2026 (ActiNet)","method_short":"ActiNet","entry_slug":"actinet-ssl-harnet-hmm-activity-intensity-sci-rep-2026","entry_kind":"paper","value":0.661,"value_text":"0.661","protocol":"Deployed CAPTURE-24 model, wrist ACC; RF+HMM 0.531.","open_weights":null}]},{"id":"har__wisdm__accuracy","dataset":"wisdm","dataset_label":"WISDM AR v1.1","metric":"accuracy","metric_label":"Accuracy","higher_is_better":true,"rows":[{"rank":1,"method":"Multilayer perceptron","method_short":"MLP (Kwapisz)","entry_slug":null,"entry_kind":null,"value":91.7,"value_text":"91.7","protocol":"WISDM v1.1 (original release), 29 users, phone in front pocket, 10 s windows, 43 features; 10-fold cross-validation over windows, NOT subject-independent; overall accuracy in %.","open_weights":null},{"rank":2,"method":"J48 decision tree","method_short":"J48 (Kwapisz)","entry_slug":null,"entry_kind":null,"value":85.1,"value_text":"85.1","protocol":"WISDM v1.1 (original release), 29 users, phone in front pocket, 10 s windows, 43 features; 10-fold cross-validation over windows, NOT subject-independent; overall accuracy in %.","open_weights":null},{"rank":3,"method":"Logistic regression","method_short":"Logistic regression (Kwapisz)","entry_slug":null,"entry_kind":null,"value":78.1,"value_text":"78.1","protocol":"WISDM v1.1 (original release), 29 users, phone in front pocket, 10 s windows, 43 features; 10-fold cross-validation over windows, NOT subject-independent; overall accuracy in %.","open_weights":null},{"rank":4,"method":"Majority-class baseline","method_short":"Majority class (Kwapisz)","entry_slug":null,"entry_kind":null,"value":37.2,"value_text":"37.2","protocol":"WISDM v1.1 (original release), 29 users, phone in front pocket, 10 s windows, 43 features; 10-fold cross-validation over windows, NOT subject-independent; overall accuracy in %.","open_weights":null}]},{"id":"har__uci-har__f1_har_unspecified","dataset":"uci-har","dataset_label":"UCI Human Activity Recognition Using Smartphones (UCI-HAR)","metric":"f1_har_unspecified","metric_label":"F1 on HAR activities (averaging not stated)","higher_is_better":true,"rows":[{"rank":1,"method":"Multi-channel autoencoder, multi-task (MCAE)","method_short":"MCAE (Azadi)","entry_slug":null,"entry_kind":null,"value":98.55,"value_text":"98.55","protocol":"Held-out test subjects (UCI-HAR uses the official 21/9 subject split); F1 in %, averaging (macro or weighted) not stated for the headline table (Table 7).","open_weights":null},{"rank":2,"method":"Single-task supervised baseline","method_short":"Single-task baseline (Azadi)","entry_slug":null,"entry_kind":null,"value":97.53,"value_text":"97.53","protocol":"Held-out test subjects (UCI-HAR uses the official 21/9 subject split); F1 in %, averaging (macro or weighted) not stated for the headline table (Table 7).","open_weights":null},{"rank":3,"method":"Classical autoencoder, multi-task (CAE)","method_short":"CAE (Azadi)","entry_slug":null,"entry_kind":null,"value":94.61,"value_text":"94.61","protocol":"Held-out test subjects (UCI-HAR uses the official 21/9 subject split); F1 in %, averaging (macro or weighted) not stated for the headline table (Table 7).","open_weights":null}]},{"id":"har__usc-had__f1_har_unspecified","dataset":"usc-had","dataset_label":"USC-HAD","metric":"f1_har_unspecified","metric_label":"F1 on HAR activities (averaging not stated)","higher_is_better":true,"rows":[{"rank":1,"method":"Multi-channel autoencoder, multi-task (MCAE)","method_short":"MCAE (Azadi)","entry_slug":null,"entry_kind":null,"value":83.9,"value_text":"83.9","protocol":"Held-out test subjects (UCI-HAR uses the official 21/9 subject split); F1 in %, averaging (macro or weighted) not stated for the headline table (Table 7).","open_weights":null},{"rank":2,"method":"Single-task supervised baseline","method_short":"Single-task baseline (Azadi)","entry_slug":null,"entry_kind":null,"value":76.17,"value_text":"76.17","protocol":"Held-out test subjects (UCI-HAR uses the official 21/9 subject split); F1 in %, averaging (macro or weighted) not stated for the headline table (Table 7).","open_weights":null},{"rank":3,"method":"Classical autoencoder, multi-task (CAE)","method_short":"CAE (Azadi)","entry_slug":null,"entry_kind":null,"value":75.69,"value_text":"75.69","protocol":"Held-out test subjects (UCI-HAR uses the official 21/9 subject split); F1 in %, averaging (macro or weighted) not stated for the headline table (Table 7).","open_weights":null}]},{"id":"har__deo__balanced_accuracy","dataset":"deo","dataset_label":"DEO","metric":"balanced_accuracy","metric_label":"Balanced accuracy (HAR / class-imbalance)","higher_is_better":true,"rows":[{"rank":1,"method":"Diffusion-Based Synthetic Data Pretraining for Enhancing Activity Recognition","method_short":"Diffusion-Based Synthetic Data Pretraining for E","entry_slug":"diffusion-synthetic-pretraining-har-2026","entry_kind":"paper","value":90.6,"value_text":"90.6","protocol":"CABiGRU pre-trained on diffusion-generated smartwatch IMU windows, fine-tuned on real data","open_weights":false}]},{"id":"har__har-6bench-avg__macro_f1","dataset":"har-6bench-avg","dataset_label":"har-6bench-avg","metric":"macro_f1","metric_label":"Macro-averaged F1","higher_is_better":true,"rows":[{"rank":1,"method":"bio-pm","method_short":"Bio-PM","entry_slug":"bio-pm","entry_kind":"model","value":0.65,"value_text":"0.65","protocol":"Average across 6 subject-disjoint benches — not a single-benchmark rank","open_weights":null}]},{"id":"har__hhar__macro_f1_har","dataset":"hhar","dataset_label":"HHAR","metric":"macro_f1_har","metric_label":"Macro-F1 on HAR activities","higher_is_better":true,"rows":[{"rank":1,"method":"inrtlwrist (InrtlWrist MAE ViT 21.4M, MIT weights)","method_short":"InrtlWrist","entry_slug":"inrtlwrist","entry_kind":"model","value":0.621,"value_text":"0.621 probe / 0.603 FT","protocol":"IMU-Bench task card: 6 classes, 2.56 s, subject-disjoint; LIMU-BERT/SimMTM/FOCAL/RelCon/SSL-Wearables retrained on the same corpus in released results.","open_weights":null}]},{"id":"har__mhealth__accuracy","dataset":"mhealth","dataset_label":"mhealth","metric":"accuracy","metric_label":"Accuracy","higher_is_better":true,"rows":[{"rank":1,"method":"Random-forest strategy (Yang)","method_short":"Random forest (Yang)","entry_slug":null,"entry_kind":null,"value":95.09,"value_text":"95.09","protocol":"Leave-one-subject-out; random-forest-based strategy; the paper reports an average F1 and accuracy in %.","open_weights":null}]},{"id":"har__opentslm-har-cot__macro_f1_har","dataset":"opentslm-har-cot","dataset_label":"opentslm-har-cot","metric":"macro_f1_har","metric_label":"Macro-F1 on HAR activities","higher_is_better":true,"rows":[{"rank":1,"method":"lapras-latent-reasoning-time-series-language-models-2026 (Lapras on ChatTS-7B)","method_short":"Lapras","entry_slug":"lapras-latent-reasoning-time-series-language-models-2026","entry_kind":"paper","value":73.53,"value_text":"73.53 (%)","protocol":"Multiple-choice HAR QA (OpenTSLM, Langer et al. 2026), 8,222 test items; same-backbone Explicit-CoT 67.27, No-CoT 63.41; OpenTSLM-1B+Lapras 68.20, SLIP-1B+Lapras 70.75.","open_weights":null}]},{"id":"har__pamap2__accuracy","dataset":"pamap2","dataset_label":"PAMAP2","metric":"accuracy","metric_label":"Accuracy","higher_is_better":true,"rows":[{"rank":1,"method":"Random-forest strategy (Yang)","method_short":"Random forest (Yang)","entry_slug":null,"entry_kind":null,"value":92.21,"value_text":"92.21","protocol":"Leave-one-subject-out; random-forest-based strategy; the paper reports an average F1 and accuracy in %.","open_weights":null}]},{"id":"har__prosi__macro_f1_har","dataset":"prosi","dataset_label":"prosi","metric":"macro_f1_har","metric_label":"Macro-F1 on HAR activities","higher_is_better":true,"rows":[{"rank":1,"method":"take-it-personally-rl-ppg-ssl-emotion-acii-2026 (RL-PPG fine-tuned)","method_short":"Take it Personally","entry_slug":"take-it-personally-rl-ppg-ssl-emotion-acii-2026","entry_kind":"paper","value":0.64,"value_text":"0.64","protocol":"7 activities, LOSO, PPG-only (Galaxy Watch 3, 25 Hz, 10 s); uniform baseline 0.14","open_weights":null}]},{"id":"har__udpad__accuracy","dataset":"udpad","dataset_label":"udpad","metric":"accuracy","metric_label":"Accuracy","higher_is_better":true,"rows":[{"rank":1,"method":"μDPad: A Large-Scale Multimodal PPG and IMU Dataset for Wrist-Worn Microgesture Recognition","method_short":"μDPad","entry_slug":"udpad-wrist-ppg-imu-microgesture-2026","entry_kind":"paper","value":83.7,"value_text":"83.7","protocol":"LOSO, 60 participants, offline window classification, 6 microgestures + unknown; μDPad intermediate-fusion VGG-style 1D CNN (670k params), Acc+Gyro+PPG(green+IR); value_secondary = F1 77.0. Online HMM: 78.6% acc, 5.3% FPR (abstract states 3.7% FPR)","open_weights":false}]},{"id":"har__wisdm__macro_f1_har","dataset":"wisdm","dataset_label":"WISDM AR v1.1","metric":"macro_f1_har","metric_label":"Macro-F1 on HAR activities","higher_is_better":true,"rows":[{"rank":1,"method":"actinet-ssl-harnet-hmm-activity-intensity-sci-rep-2026 (ActiNet)","method_short":"ActiNet","entry_slug":"actinet-ssl-harnet-hmm-activity-intensity-sci-rep-2026","entry_kind":"paper","value":0.741,"value_text":"0.741","protocol":"Deployed CAPTURE-24 model, labels mapped to intensity; RF+HMM 0.679.","open_weights":null}]}]},{"task":"cuffless-bp","task_name":"Cuffless BP","total_rows":24,"boards":[{"id":"cuffless-bp__ppgbp-guilin__mae_mmhg","dataset":"ppgbp-guilin","dataset_label":"PPG-BP Database (Guilin / Liang 2018)","metric":"mae_mmhg","metric_label":"Mean absolute error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"Chowdhury GPR + ReliefF (Sensors 2020)","method_short":"Chowdhury GPR + ReliefF","entry_slug":"chowdhury-ppg-demographics-ml-bp-sensors-2020","entry_kind":"paper","value":3.02,"value_text":"3.02","protocol":"Record-level random 85/15 split — NOT subject-disjoint; demographics in features; SBP fails AAMI SD","open_weights":false},{"rank":2,"method":"TF-MambaNet","method_short":"TF-MambaNet","entry_slug":"tf-mambanet-bidirectional-mamba-ppg-fm-icassp-2026","entry_kind":"paper","value":12.89,"value_text":"12.89","protocol":"PPG-BP (Liang 2018) 2 s segments, frozen features (paper Table 1); finger clinical PPG; DBP Chronos 8.37 better","open_weights":false},{"rank":3,"method":"TEX-MEX 0.56M PPG FM (linear probe)","method_short":"TEX-MEX 0.56M PPG FM","entry_slug":"tex-mex","entry_kind":"model","value":14.007,"value_text":"14.007","protocol":"Frozen LP, subject-level 5-fold CV × 15 seeds (arXiv 2610.19557 Table 4); same-paper Pulse-PPG 16.62/9.65, PaPaGei-P 15.47/9.15; finger clinical; weights promised","open_weights":false}]},{"id":"cuffless-bp__uci-bp__mae_mmhg","dataset":"uci-bp","dataset_label":"uci-bp","metric":"mae_mmhg","metric_label":"Mean absolute error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"ExpertoRhythm","method_short":"ExpertoRhythm","entry_slug":"expertorhythm-morphology-aware-ppg-abp-bp-2026","entry_kind":"paper","value":2.46,"value_text":"2.46","protocol":"UCI cuff-less BP 942 subj; 10 s @ 125 Hz; lab PPG","open_weights":false},{"rank":2,"method":"Hybrid features + personalized calibration (Zhang, Sensors 2026)","method_short":"Hybrid features + personalized calibration","entry_slug":"personalized-calibration-hybrid-ppg-ecg-bp-sensors-2026","entry_kind":"paper","value":7.85,"value_text":"7.85","protocol":"PPG+ECG, base-BP calibration; split unstated (abstract)","open_weights":false},{"rank":3,"method":"ECG-PPG JEPA 345M (linear probe)","method_short":"ECG-PPG JEPA 345M","entry_slug":"ecg-ppg-jepa-latent-space-prediction-kdd-2026","entry_kind":"paper","value":11.94,"value_text":"11.94","protocol":"Nested 5-fold CV (AnyPPG protocol); clinical finger","open_weights":false}]},{"id":"cuffless-bp__pulsedb__mae_mmhg","dataset":"pulsedb","dataset_label":"PulseDB v2.0","metric":"mae_mmhg","metric_label":"Mean absolute error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"PPG-FusionNet (Computers 2026)","method_short":"PPG-FusionNet","entry_slug":"ppg-fusionnet-dual-branch-cuffless-bp-computers-2026","entry_kind":"paper","value":7.46,"value_text":"7.46","protocol":"PPG-only; PulseDB subset (CalBased/CalFree) not stated in abstract","open_weights":false},{"rank":2,"method":"GPT-PPG: a GPT-based foundation model for photoplethysmography signals","method_short":"GPT-PPG","entry_slug":"gpt-ppg-gpt-foundation-model-ppg-physiolmeas-2025","entry_kind":"paper","value":8.12,"value_text":"8.12","protocol":"SBP/DBP MAE, GPT-1B fine-tuned; authors' random subject-disjoint test set (not official PulseDB subsets); authors call it reference only; arXiv v1 Table 4","open_weights":false}]},{"id":"cuffless-bp__aurora-bp__mae_mmhg","dataset":"aurora-bp","dataset_label":"aurora-bp","metric":"mae_mmhg","metric_label":"Mean absolute error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"PulseAnchor (BSN 2026)","method_short":"PulseAnchor","entry_slug":"pulseanchor-personalizing-ppg-fm-bp-bsn-2026","entry_kind":"paper","value":7.84,"value_text":"7.84","protocol":"Wrist PPG; FM + subject calibration examples (relative prediction); calibration count unstated","open_weights":false}]},{"id":"cuffless-bp__aurora-bp-htn__accuracy","dataset":"aurora-bp-htn","dataset_label":"aurora-bp-htn","metric":"accuracy","metric_label":"Accuracy","higher_is_better":true,"rows":[{"rank":1,"method":"Awake–sleep PPG indices + SVM (EHJ-DH 2026)","method_short":"Awake–sleep PPG indices + SVM","entry_slug":"hypertension-awake-sleep-ppg-aurora-ehjdh-2026","entry_kind":"paper","value":0.811,"value_text":"0.811","protocol":"Hypertension yes/no, subject-level, 18 held-out subj; F1 0.828","open_weights":false}]},{"id":"cuffless-bp__cuhk-bp-htn__accuracy","dataset":"cuhk-bp-htn","dataset_label":"cuhk-bp-htn","metric":"accuracy","metric_label":"Accuracy","higher_is_better":true,"rows":[{"rank":1,"method":"Awake–sleep PPG temporal indices + SVM (external)","method_short":"Awake–sleep PPG temporal indices + SVM","entry_slug":"hypertension-awake-sleep-ppg-aurora-ehjdh-2026","entry_kind":"paper","value":0.846,"value_text":"0.846","protocol":"External CUHK-BP n=26, no retraining; F1 0.867","open_weights":false}]},{"id":"cuffless-bp__liu-wrist-9ch-array-121__mae_mmhg","dataset":"liu-wrist-9ch-array-121","dataset_label":"liu-wrist-9ch-array-121","metric":"mae_mmhg","metric_label":"Mean absolute error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"3M-BPNet (Liu, JBHI)","method_short":"3M-BPNet","entry_slug":"liu-3m-bpnet-multi-site-wrist-ppg-bp-jbhi-2025","entry_kind":"paper","value":4.84,"value_text":"4.84","protocol":"Wrist 9-ch single-λ array, own 121 subj, personalized calibration; best 5-ch C5–C9 1.33/1.16; abstract-only, split unknown","open_weights":false}]},{"id":"cuffless-bp__mcbp-cold-pressor-23__mae_mmhg","dataset":"mcbp-cold-pressor-23","dataset_label":"mcbp-cold-pressor-23","metric":"mae_mmhg","metric_label":"Mean absolute error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"McBP-Net (JBHI 2025)","method_short":"McBP-Net","entry_slug":"mcbp-net-beat-to-beat-bp-multimodal-cnn-lstm-jbhi-2025","entry_kind":"paper","value":4.19,"value_text":"4.19","protocol":"Own 23 subj, cold pressor, beat-to-beat; PPG+ECG+IPG+skin temp; abstract-only","open_weights":false}]},{"id":"cuffless-bp__mimic-bp__sde_mmhg","dataset":"mimic-bp","dataset_label":"MIMIC-BP","metric":"sde_mmhg","metric_label":"Standard deviation of error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"FiLMamba","method_short":"FiLMamba","entry_slug":"filmamba-film-conditioned-mamba-cuffless-bp-2026","entry_kind":"paper","value":6.45,"value_text":"6.45","protocol":"MIMIC-BP SDE SBP/DBP; clinical/lab PPG","open_weights":false}]},{"id":"cuffless-bp__mimic-waveform-205k__mae_mmhg","dataset":"mimic-waveform-205k","dataset_label":"mimic-waveform-205k","metric":"mae_mmhg","metric_label":"Mean absolute error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"Multiresolution U-Net @62.5 Hz (Physiol Meas 2026)","method_short":"Multiresolution U-Net @62.5 Hz","entry_slug":"practical-optimization-dl-ppg-bp-sampling-physiolmeas-2026","entry_kind":"paper","value":5.24,"value_text":"5.24","protocol":"Segment-level (authors: subject-level validation still needed); 125 Hz = 6.34/3.27","open_weights":false}]},{"id":"cuffless-bp__ppg-bp__mae_mmhg","dataset":"ppg-bp","dataset_label":"ppg-bp","metric":"mae_mmhg","metric_label":"Mean absolute error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"PPG-Bench: A Unified Benchmark for PPG Foundation Models","method_short":"PPG-Bench","entry_slug":"ppg-bench-unified-benchmark-ppg-foundation-models-2026","entry_kind":"paper","value":13.32,"value_text":"13.32","protocol":"PPG-Bench linear probe SBP/DBP MAE; AnyPPG; subject-level; subject-disjoint 5-fold","open_weights":false}]},{"id":"cuffless-bp__ppgbp-guilin__auroc","dataset":"ppgbp-guilin","dataset_label":"PPG-BP Database (Guilin / Liang 2018)","metric":"auroc","metric_label":"Area under ROC","higher_is_better":true,"rows":[{"rank":1,"method":"TEX-MEX 0.56M PPG FM (linear probe)","method_short":"TEX-MEX 0.56M PPG FM","entry_slug":"tex-mex","entry_kind":"model","value":0.7347,"value_text":"0.7347","protocol":"Hypertension class, PRAUC 0.4960; next best MOMENT 0.6749 / AnyPPG 0.6739 (same paper); classification — not comparable to MAE rows","open_weights":false}]},{"id":"cuffless-bp__pulsedb-calbased__mae_mmhg","dataset":"pulsedb-calbased","dataset_label":"pulsedb-calbased","metric":"mae_mmhg","metric_label":"Mean absolute error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"Dilated-residual attention U-Net (Information 2026)","method_short":"Dilated-residual attention U-Net","entry_slug":"dilated-residual-attention-unet-calibrated-bp-information-2026","entry_kind":"paper","value":3.81,"value_text":"3.81","protocol":"PulseDB calibration-based subset (same subjects train/test); PPG+ECG","open_weights":false}]},{"id":"cuffless-bp__pulsedb-vital__mae_mmhg","dataset":"pulsedb-vital","dataset_label":"pulsedb-vital","metric":"mae_mmhg","metric_label":"Mean absolute error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"ECG-PPG JEPA 345M (linear probe)","method_short":"ECG-PPG JEPA 345M","entry_slug":"ecg-ppg-jepa-latent-space-prediction-kdd-2026","entry_kind":"paper","value":8.23,"value_text":"8.23","protocol":"Vital-BP predefined split from PulseDB; PulseDB also in pretraining (overlap not discussed); clinical finger","open_weights":false}]},{"id":"cuffless-bp__pulsedb-vitaldb-aha6__accuracy","dataset":"pulsedb-vitaldb-aha6","dataset_label":"pulsedb-vitaldb-aha6","metric":"accuracy","metric_label":"Accuracy","higher_is_better":true,"rows":[{"rank":1,"method":"Few-shot on frozen FM embeddings (BSN 2026)","method_short":"Few-shot on frozen FM embeddings","entry_slug":"few-shot-fm-embeddings-cuffless-bp-aha-bsn-2026","entry_kind":"paper","value":0.824,"value_text":"0.824","protocol":"AHA 6-class category, 3-shot; weighted F1 0.832; classification — not comparable to MAE rows","open_weights":false}]},{"id":"cuffless-bp__ring-hgcpt__sde_mmhg","dataset":"ring-hgcpt","dataset_label":"ring-hgcpt","metric":"sde_mmhg","metric_label":"Standard deviation of error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"DPIKAN (ICLR'27 sub)","method_short":"DPIKAN","entry_slug":"dpikan-physics-informed-kan-cuffless-bp-ring-iclr2027-sub","entry_kind":"paper","value":5.41,"value_text":"5.41","protocol":"Smart-ring PPG, controlled CV perturbations; LOSO; ME 1.08/0.09","open_weights":false}]},{"id":"cuffless-bp__uci-bp__sde_mmhg","dataset":"uci-bp","dataset_label":"uci-bp","metric":"sde_mmhg","metric_label":"Standard deviation of error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"FiLMamba (train-free transfer)","method_short":"FiLMamba","entry_slug":"filmamba-film-conditioned-mamba-cuffless-bp-2026","entry_kind":"paper","value":4.29,"value_text":"4.29","protocol":"UCI-BP transfer SDE; clinical/lab","open_weights":false}]},{"id":"cuffless-bp__vitaldb__mae_mmhg","dataset":"vitaldb","dataset_label":"VitalDB","metric":"mae_mmhg","metric_label":"Mean absolute error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"TCN–BiLSTM PPG→ABP (AHJ Plus 2026)","method_short":"TCN–BiLSTM PPG→ABP","entry_slug":"tcn-bilstm-ppg-to-abp-vitaldb-ahjo-2026","entry_kind":"paper","value":3.61,"value_text":"3.61","protocol":"2,226 patients; within-VitalDB split unstated; MIMIC-IV external (no abstract numbers)","open_weights":false}]},{"id":"cuffless-bp__windkessel-own-10__mae_mmhg","dataset":"windkessel-own-10","dataset_label":"windkessel-own-10","metric":"mae_mmhg","metric_label":"Mean absolute error (mmHg); cuffless BP","higher_is_better":false,"rows":[{"rank":1,"method":"Windkessel-regularized (IEEE Sens Lett 2026)","method_short":"Windkessel-regularized","entry_slug":"windkessel-regularized-ppg-bp-lsens-2026","entry_kind":"paper","value":4.1,"value_text":"4.10","protocol":"Own 10 healthy subj, finger wearable; split unstated","open_weights":false}]}]},{"task":"hr-estimation","task_name":"Heart rate estimation","total_rows":22,"boards":[{"id":"hr-estimation__ppg-dalia__mae_bpm","dataset":"ppg-dalia","dataset_label":"PPG-DaLiA","metric":"mae_bpm","metric_label":"Mean absolute HR error (BPM)","higher_is_better":false,"rows":[{"rank":1,"method":"CardioState-JEPA","method_short":"CardioState-JEPA","entry_slug":"cardiostate-jepa","entry_kind":"model","value":3.8,"value_text":"3.8","protocol":"Paper Table 1 frozen","open_weights":"HF listed"},{"rank":2,"method":"GPT-PPG: a GPT-based foundation model for photoplethysmography signals","method_short":"GPT-PPG","entry_slug":"gpt-ppg-gpt-foundation-model-ppg-physiolmeas-2025","entry_kind":"paper","value":4.77,"value_text":"4.77","protocol":"GPT-1B fine-tuned (MOFT), LOSO; arXiv v1 Table 2; not comparable to frozen-probe rows","open_weights":false},{"rank":3,"method":"ECG-PPG JEPA 345M (linear probe)","method_short":"ECG-PPG JEPA 345M","entry_slug":"ecg-ppg-jepa-latent-space-prediction-kdd-2026","entry_kind":"paper","value":4.96,"value_text":"4.96","protocol":"LOSO, frozen; PPG-only input","open_weights":false},{"rank":4,"method":"TF-MambaNet","method_short":"TF-MambaNet","entry_slug":"tf-mambanet-bidirectional-mamba-ppg-fm-icassp-2026","entry_kind":"paper","value":7.27,"value_text":"7.27","protocol":"8 s windows, frozen features (paper Table 1); split unstated; same-paper re-runs NormWear 10.16 / PaPaGei 12.25 / Chronos 12.34","open_weights":false},{"rank":5,"method":"PPG-Bench: A Unified Benchmark for PPG Foundation Models","method_short":"PPG-Bench","entry_slug":"ppg-bench-unified-benchmark-ppg-foundation-models-2026","entry_kind":"paper","value":7.39,"value_text":"7.39","protocol":"PPG-Bench linear probe; AnyPPG; subject-disjoint 5-fold; mean±95% CI in paper","open_weights":false},{"rank":6,"method":"HiPPT (linear probe)","method_short":"HiPPT","entry_slug":"hippt","entry_kind":"model","value":7.65,"value_text":"7.65","protocol":"15-subject LOSO, frozen; 12-s segments → 250 Hz; same-paper re-runs Apple 8.18 / PaPaGei 8.29; 2026-10-14","open_weights":"false (promised)"},{"rank":7,"method":"Pulse-PPG","method_short":"Pulse-PPG","entry_slug":"pulse-ppg","entry_kind":"model","value":8.936,"value_text":"8.936","protocol":"Frozen linear-probe; paper protocol","open_weights":"true (MIT)"},{"rank":8,"method":"MSMA-PPG (PPG-only)","method_short":"MSMA-PPG","entry_slug":"msma-ppg","entry_kind":"model","value":9.246,"value_text":"9.246","protocol":"PPG-only — not comparable to PPG+ACC SOTA rows above; ICLR'27 sub","open_weights":null},{"rank":9,"method":"AnyPPG","method_short":"AnyPPG","entry_slug":"anyppg","entry_kind":"model","value":9.27,"value_text":"9.27","protocol":"Paper linear-probe","open_weights":"yes / license unclear"},{"rank":10,"method":"TEX-MEX 0.56M (linear probe)","method_short":"TEX-MEX 0.56M","entry_slug":"tex-mex","entry_kind":"model","value":9.79,"value_text":"9.790","protocol":"Subject-level 5-fold CV × 15 seeds (arXiv 2610.19557 Table 4); same-paper Pulse-PPG 10.105, AnyPPG 6.643; not Pulse-PPG-paper protocol; 2026-10-15","open_weights":"false (promised)"},{"rank":11,"method":"PaPaGei-S (re-eval)","method_short":"PaPaGei-S","entry_slug":"papagei-s","entry_kind":"model","value":15.22,"value_text":"15.22","protocol":"Under Pulse-PPG protocol (not PaPaGei original)","open_weights":"true (BSD-3)"}]},{"id":"hr-estimation__ieee-spc-2015-troika__mae_bpm","dataset":"ieee-spc-2015-troika","dataset_label":"IEEE Signal Processing Cup 2015 (TROIKA)","metric":"mae_bpm","metric_label":"Mean absolute HR error (BPM)","higher_is_better":false,"rows":[{"rank":1,"method":"GPT-PPG: a GPT-based foundation model for photoplethysmography signals","method_short":"GPT-PPG","entry_slug":"gpt-ppg-gpt-foundation-model-ppg-physiolmeas-2025","entry_kind":"paper","value":1.98,"value_text":"1.98","protocol":"GPT-1B fine-tuned (MOFT), LOSO; arXiv v1 Table 2","open_weights":false},{"rank":2,"method":"ECG-PPG JEPA 345M (linear probe)","method_short":"ECG-PPG JEPA 345M","entry_slug":"ecg-ppg-jepa-latent-space-prediction-kdd-2026","entry_kind":"paper","value":2.09,"value_text":"2.09","protocol":"LOSO, frozen","open_weights":false}]},{"id":"hr-estimation__wesad__mae_bpm","dataset":"wesad","dataset_label":"WESAD","metric":"mae_bpm","metric_label":"Mean absolute HR error (BPM)","higher_is_better":false,"rows":[{"rank":1,"method":"ECG-PPG JEPA 345M (linear probe)","method_short":"ECG-PPG JEPA 345M","entry_slug":"ecg-ppg-jepa-latent-space-prediction-kdd-2026","entry_kind":"paper","value":4.58,"value_text":"4.58","protocol":"LOSO, frozen encoder (KDD'26 Table 2)","open_weights":false},{"rank":2,"method":"GPT-PPG: a GPT-based foundation model for photoplethysmography signals","method_short":"GPT-PPG","entry_slug":"gpt-ppg-gpt-foundation-model-ppg-physiolmeas-2025","entry_kind":"paper","value":4.98,"value_text":"4.98","protocol":"GPT-1B fine-tuned (MOFT), LOSO; arXiv v1 Table 2","open_weights":false}]},{"id":"hr-estimation__bilateral-sports-30__mae_bpm","dataset":"bilateral-sports-30","dataset_label":"bilateral-sports-30","metric":"mae_bpm","metric_label":"Mean absolute HR error (BPM)","higher_is_better":false,"rows":[{"rank":1,"method":"Bilateral-PPG → single-wrist KD student (SD4H 2026)","method_short":"Bilateral-PPG → single-wrist KD student","entry_slug":"bilateral-ppg-cross-site-attention-kd-hr-sd4h-2026","entry_kind":"paper","value":2.79,"value_text":"2.79","protocol":"Own 30 subj, 6 sports, ECG truth; non-dominant wrist; dominant 3.72; bilateral teacher 2.85","open_weights":false}]},{"id":"hr-estimation__butppg__mae_bpm","dataset":"butppg","dataset_label":"Brno University of Technology Smartphone PPG Database (BUT PPG)","metric":"mae_bpm","metric_label":"Mean absolute HR error (BPM)","higher_is_better":false,"rows":[{"rank":1,"method":"ECG-PPG JEPA 345M (linear probe)","method_short":"ECG-PPG JEPA 345M","entry_slug":"ecg-ppg-jepa-latent-space-prediction-kdd-2026","entry_kind":"paper","value":7.86,"value_text":"7.86","protocol":"Nested 5-fold CV (AnyPPG protocol); smartphone PPG","open_weights":false}]},{"id":"hr-estimation__earset__mae_bpm","dataset":"earset","dataset_label":"EarSet","metric":"mae_bpm","metric_label":"Mean absolute HR error (BPM)","higher_is_better":false,"rows":[{"rank":1,"method":"MSMA-PPG","method_short":"MSMA-PPG","entry_slug":"msma-ppg","entry_kind":"model","value":5.02,"value_text":"5.020","protocol":"PPG-only single selected λ; participant-macro; routing off; ICLR'27 sub","open_weights":null}]},{"id":"hr-estimation__gyro-acc-ppg-bami1__mae_bpm","dataset":"gyro-acc-ppg-bami1","dataset_label":"Gyro-Acc-PPG (BAMI-I, Wonkwang)","metric":"mae_bpm","metric_label":"Mean absolute HR error (BPM)","higher_is_better":false,"rows":[{"rank":1,"method":"ECG-PPG JEPA 345M (linear probe)","method_short":"ECG-PPG JEPA 345M","entry_slug":"ecg-ppg-jepa-latent-space-prediction-kdd-2026","entry_kind":"paper","value":10.22,"value_text":"10.22","protocol":"Nested 5-fold CV (AnyPPG protocol); treadmill wrist PPG","open_weights":false}]},{"id":"hr-estimation__ppg-bp__mae_bpm","dataset":"ppg-bp","dataset_label":"ppg-bp","metric":"mae_bpm","metric_label":"Mean absolute HR error (BPM)","higher_is_better":false,"rows":[{"rank":1,"method":"PPG-Bench: A Unified Benchmark for PPG Foundation Models","method_short":"PPG-Bench","entry_slug":"ppg-bench-unified-benchmark-ppg-foundation-models-2026","entry_kind":"paper","value":3.31,"value_text":"3.31","protocol":"PPG-Bench linear probe; AnyPPG; subject-level aggregation; subject-disjoint 5-fold","open_weights":false}]},{"id":"hr-estimation__ppg-dalia__exact_match","dataset":"ppg-dalia","dataset_label":"PPG-DaLiA","metric":"exact_match","metric_label":null,"higher_is_better":true,"rows":[{"rank":1,"method":"PulseLM (Qwen2.5-7B + Pulse-PPG)","method_short":"PulseLM","entry_slug":"pulselm-ppg-text-qa-2026","entry_kind":"paper","value":0.862,"value_text":"0.862","protocol":"10 s @125 Hz, subject-disjoint ~8/1/1 per source; categorical QA exact-match (arXiv 2603.03331 Table 2); HR 3-class (<60/60-100/>100 bpm); no MAE reported, cannot convert; Blind 0.837","open_weights":"HF checkpoint linked; code CC BY 4.0"}]},{"id":"hr-estimation__wildppg__mae_bpm","dataset":"wildppg","dataset_label":"WildPPG","metric":"mae_bpm","metric_label":"Mean absolute HR error (BPM)","higher_is_better":false,"rows":[{"rank":1,"method":"PPG-Bench: A Unified Benchmark for PPG Foundation Models","method_short":"PPG-Bench","entry_slug":"ppg-bench-unified-benchmark-ppg-foundation-models-2026","entry_kind":"paper","value":9.76,"value_text":"9.76","protocol":"PPG-Bench linear probe; AnyPPG; subject-disjoint 5-fold","open_weights":false}]}]},{"task":"afib","task_name":"Atrial fibrillation detection","total_rows":14,"boards":[{"id":"afib__stanford-ppg__auroc","dataset":"stanford-ppg","dataset_label":"Stanford wrist PPG AF (DeepBeat)","metric":"auroc","metric_label":"Area under ROC","higher_is_better":true,"rows":[{"rank":1,"method":"xECG (FT; ECG-pretrained xLSTM)","method_short":"xECG","entry_slug":"benchecg-xecg-xlstm-ecg-fm-2026","entry_kind":"paper","value":0.883,"value_text":"0.883","protocol":"DeepBeat wrist-PPG AF, full finetune, 5-run mean ±0.019 (npj 2026 BenchECG); same protocol ECGFounder 0.846, ECG-CPC 0.832","open_weights":"true (MIT)"},{"rank":2,"method":"xECG (linear probe)","method_short":"xECG","entry_slug":"benchecg-xecg-xlstm-ecg-fm-2026","entry_kind":"paper","value":0.751,"value_text":"0.751","protocol":"Frozen ECG-pretrained encoder on wrist PPG, 5-run mean ±0.013","open_weights":"true (MIT)"}]},{"id":"afib__afppgecg__balanced_accuracy","dataset":"afppgecg","dataset_label":"Long-term ECG and wrist PPG with annotated AF episodes (AFPPGECG)","metric":"balanced_accuracy","metric_label":"Balanced accuracy (HAR / class-imbalance)","higher_is_better":true,"rows":[{"rank":1,"method":"VitalAgent (tool-feature AF screen; arXiv 2605.29483)","method_short":"VitalAgent","entry_slug":"vitalagent-tool-augmented-ecg-ppg-monitoring-agent-2026","entry_kind":"paper","value":0.672,"value_text":"0.672","protocol":"Per-window 10 s, deterministic RR-variability tools, no training; sens 78.3 / spec 56.1; 44 pts; only row on this axis","open_weights":"false (code MIT, no weights)"}]},{"id":"afib__afppgecg__exact_match","dataset":"afppgecg","dataset_label":"Long-term ECG and wrist PPG with annotated AF episodes (AFPPGECG)","metric":"exact_match","metric_label":null,"higher_is_better":true,"rows":[{"rank":1,"method":"PulseLM (best: Qwen3-4B / LLaMA + PaPaGei)","method_short":"PulseLM","entry_slug":"pulselm-ppg-text-qa-2026","entry_kind":"paper","value":0.861,"value_text":"0.861","protocol":"10 s @125 Hz, subject-disjoint ~8/1/1 per source; categorical QA exact-match (arXiv 2603.03331 Table 2); text-only Blind also 0.861, so no gain over label prior; not balanced accuracy, not comparable to VitalAgent 0.672","open_weights":"HF checkpoint linked; code CC BY 4.0"}]},{"id":"afib__external-single-lead-af__auroc","dataset":"external-single-lead-af","dataset_label":"external-single-lead-af","metric":"auroc","metric_label":"Area under ROC","higher_is_better":true,"rows":[{"rank":1,"method":"ECGFounder","method_short":"ECGFounder","entry_slug":"ecgfounder","entry_kind":"model","value":0.957,"value_text":"0.957","protocol":"External single-lead AF","open_weights":"true (MIT)"}]},{"id":"afib__icentia11k__balanced_accuracy","dataset":"icentia11k","dataset_label":"Icentia11k single-lead continuous raw ECG (PhysioNet)","metric":"balanced_accuracy","metric_label":"Balanced accuracy (HAR / class-imbalance)","higher_is_better":true,"rows":[{"rank":1,"method":"VitalAgent (tool-feature AF screen)","method_short":"VitalAgent","entry_slug":"vitalagent-tool-augmented-ecg-ppg-monitoring-agent-2026","entry_kind":"paper","value":0.678,"value_text":"0.678","protocol":"Same pipeline on ECG; sens 71.9 / spec 63.6; 25 pts; only row on this axis","open_weights":"false (code MIT, no weights)"}]},{"id":"afib__liu-ppg-arrhythmia-6class__auroc","dataset":"liu-ppg-arrhythmia-6class","dataset_label":"liu-ppg-arrhythmia-6class","metric":"auroc","metric_label":"Area under ROC","higher_is_better":true,"rows":[{"rank":1,"method":"Pulse2View: Same-Sensor Cross-View Pre-Training for Patient-Level Multi-Disease Classification from Wearable PPG","method_short":"Pulse2View","entry_slug":"pulse2view-same-sensor-cross-view-ppg-pretraining-iclr2027-sub","entry_kind":"paper","value":0.9697,"value_text":"0.9697","protocol":"Window-level six-class Macro AUROC (mean ± 0.0011, 3 seeds) on public Liu PPG arrhythmia dataset; multi-class rhythm, not AF-vs-rest","open_weights":false}]},{"id":"afib__mimic-iii-ext-ppg__macro_f1","dataset":"mimic-iii-ext-ppg","dataset_label":"MIMIC-III-Ext-PPG","metric":"macro_f1","metric_label":"Macro-averaged F1","higher_is_better":true,"rows":[{"rank":1,"method":"PhysioSkeptic (ICLR'27 sub)","method_short":"PhysioSkeptic","entry_slug":"physioskeptic-sqi-grounded-ecg-ppg-rhythm-reasoning-iclr2027-sub","entry_kind":"paper","value":0.886,"value_text":"0.886","protocol":"ECG–PPG multi-class rhythm; SQI-anchored LLM (GPT-5.2); not AF-only","open_weights":false}]},{"id":"afib__mimic-perform__accuracy","dataset":"mimic-perform","dataset_label":"MIMIC PERform","metric":"accuracy","metric_label":"Accuracy","higher_is_better":true,"rows":[{"rank":1,"method":"TF-MambaNet","method_short":"TF-MambaNet","entry_slug":"tf-mambanet-bidirectional-mamba-ppg-fm-icassp-2026","entry_kind":"paper","value":0.83,"value_text":"0.83","protocol":"MIMIC 8 s AF, frozen features; accuracy only (no AUROC) — not comparable to AUROC rows; Chronos 0.81 / NormWear 0.78 / PaPaGei 0.66","open_weights":false}]},{"id":"afib__mimic-perform__exact_match","dataset":"mimic-perform","dataset_label":"MIMIC PERform","metric":"exact_match","metric_label":null,"higher_is_better":true,"rows":[{"rank":1,"method":"PulseLM (LLaMA3.1-8B + Pulse-PPG)","method_short":"PulseLM","entry_slug":"pulselm-ppg-text-qa-2026","entry_kind":"paper","value":0.75,"value_text":"0.750","protocol":"10 s @125 Hz, subject-disjoint ~8/1/1 per source; categorical QA exact-match (arXiv 2603.03331 Table 2); binary AF; text-only Blind 0.488. Not comparable to TF-MambaNet 0.83 (8 s, different split)","open_weights":"HF checkpoint linked; code CC BY 4.0"}]},{"id":"afib__ptb-xl-rhythm__auroc","dataset":"ptb-xl-rhythm","dataset_label":"ptb-xl-rhythm","metric":"auroc","metric_label":"Area under ROC","higher_is_better":true,"rows":[{"rank":1,"method":"MS-ECG-FM","method_short":"MS-ECG-FM","entry_slug":"ms-ecg-fm-multi-source-contrastive-ecg-fm-2026","entry_kind":"paper","value":0.987,"value_text":"0.987","protocol":"PTB-XL Rhythm AFIB label, 12-lead, linear probe, 100% labels (Table 3 / D1); MIMIC-IV-ECG pretrain","open_weights":false}]},{"id":"afib__pulsewatch__auroc","dataset":"pulsewatch","dataset_label":"Pulsewatch","metric":"auroc","metric_label":"Area under ROC","higher_is_better":true,"rows":[{"rank":1,"method":"Development of an efficient deep learning model for arrhythmia detection from consumer smartwatches","method_short":"Development of an efficient deep learning model ","entry_slug":"george-pulsewatch-smartwatch-ppg-acc-arrhythmia-rf-vs-cnn-bilstm-frai-2026","entry_kind":"paper","value":0.9703,"value_text":"0.9703","protocol":"Same protocol; macro (one-vs-rest) ROC-AUC, Random Forest; not AF-vs-rest binary","open_weights":false}]},{"id":"afib__pulsewatch__macro_f1","dataset":"pulsewatch","dataset_label":"Pulsewatch","metric":"macro_f1","metric_label":"Macro-averaged F1","higher_is_better":true,"rows":[{"rank":1,"method":"Random Forest, engineered features (George et al., Front AI 2026)","method_short":"Random Forest, engineered features","entry_slug":"george-pulsewatch-smartwatch-ppg-acc-arrhythmia-rf-vs-cnn-bilstm-frai-2026","entry_kind":"paper","value":0.842,"value_text":"0.842","protocol":"3-class NSR/AF/PAC-PVC, 30 s, 2-fold subject-independent (72 subj); macro AUROC 0.970, acc 0.912; CNN-BiLSTM macro recall 0.874; only rows on this axis","open_weights":"false (no code)"}]},{"id":"afib__stanford-ppg__macro_f1","dataset":"stanford-ppg","dataset_label":"Stanford wrist PPG AF (DeepBeat)","metric":"macro_f1","metric_label":"Macro-averaged F1","higher_is_better":true,"rows":[{"rank":1,"method":"ECG-PPG JEPA 345M (linear probe)","method_short":"ECG-PPG JEPA 345M","entry_slug":"ecg-ppg-jepa-latent-space-prediction-kdd-2026","entry_kind":"paper","value":0.834,"value_text":"0.834","protocol":"Predefined split, frozen encoder; F1 — not comparable to AUROC rows; 11M 0.791","open_weights":false}]}]},{"task":"mwl","task_name":"Mental workload","total_rows":10,"boards":[{"id":"mwl__wesad__auroc","dataset":"wesad","dataset_label":"WESAD","metric":"auroc","metric_label":"Area under ROC","higher_is_better":true,"rows":[{"rank":1,"method":"Cross-Modal Correlation Fusion for End-to-End ECG–PPG Stress Classification","method_short":"Cross-Modal Correlation Fusion for End-to-End EC","entry_slug":"cross-modal-ecg-ppg-stress-wesad-2026","entry_kind":"paper","value":0.9729,"value_text":"0.9729","protocol":"WESAD subject-independent 5-fold CV; dual ECG+PPG encoders + cross-modal correlation fusion; paper AUC-ROC 97.29%","open_weights":false},{"rank":2,"method":"Pulse-PPG","method_short":"Pulse-PPG","entry_slug":"pulse-ppg","entry_kind":"model","value":0.9687,"value_text":"0.9687","protocol":"Binary stress; paper non-stress definition","open_weights":"true (MIT)"},{"rank":3,"method":"Thapa wrist-PPG features (LR, all features; Stage 2 exploratory)","method_short":"Thapa wrist-PPG features","entry_slug":"thapa-wrist-ppg-stress-wesad-interval-vs-morphology-2026","entry_kind":"paper","value":0.965,"value_text":"0.965","protocol":"LOSO pooled AUC, BVP 64 Hz 60 s windows, stress from 9 subjects; post hoc; interval-only 0.911–0.933; RF 0.972 transductive z-scoring (not ranked)","open_weights":"n/a (no code)"},{"rank":4,"method":"TEX-MEX 0.56M (linear probe)","method_short":"TEX-MEX 0.56M","entry_slug":"tex-mex","entry_kind":"model","value":0.963,"value_text":"0.9630","protocol":"Binary stress, subject-level 5-fold CV × 15 seeds (arXiv 2610.19557 Table 4); same-paper Pulse-PPG 0.9388 / AnyPPG 0.9520 — different protocol from Pulse-PPG paper row","open_weights":"false (promised)"},{"rank":5,"method":"Quality-aware PPG+ACC (Zenodo 22858828)","method_short":"Quality-aware PPG+ACC","entry_slug":"quality-aware-wrist-ppg-stress-wesad-zenodo-2026","entry_kind":"paper","value":0.956,"value_text":"0.956","protocol":"LOSO 15 subj, 60-s windows, baseline vs stress; low-tier venue; selective prediction improves accepted-window F1","open_weights":null},{"rank":6,"method":"Calibrated photoplethysmography-derived heart rate variability approximates electrocardiographic gold-standard performance for stress-state classification","method_short":"Calibrated photoplethysmography-derived heart ra","entry_slug":"calibrated-ppg-hrv-ecg-stress-aud-relapse-2026","entry_kind":"paper","value":0.85,"value_text":"0.85","protocol":"binary stress; classical ML on HRV features; PPG features linearly calibrated to ECG space (uncalibrated PPG 0.79; ECG 0.85); 15 subjects","open_weights":false},{"rank":7,"method":"PaPaGei-S (re-eval)","method_short":"PaPaGei-S","entry_slug":"papagei-s","entry_kind":"model","value":0.8043,"value_text":"0.8043","protocol":"Under Pulse-PPG protocol","open_weights":"true (BSD-3)"}]},{"id":"mwl__wesad__accuracy","dataset":"wesad","dataset_label":"WESAD","metric":"accuracy","metric_label":"Accuracy","higher_is_better":true,"rows":[{"rank":1,"method":"Cross-Modal Correlation Fusion for End-to-End ECG–PPG Stress Classification","method_short":"Cross-Modal Correlation Fusion for End-to-End EC","entry_slug":"cross-modal-ecg-ppg-stress-wesad-2026","entry_kind":"paper","value":0.9034,"value_text":"0.9034","protocol":"WESAD subject-independent 5-fold CV; mean Accuracy 90.34%; Stress Recall 94.07%","open_weights":false},{"rank":2,"method":"TF-MambaNet","method_short":"TF-MambaNet","entry_slug":"tf-mambanet-bidirectional-mamba-ppg-fm-icassp-2026","entry_kind":"paper","value":0.81,"value_text":"0.81","protocol":"Binary stress, 8 s, frozen features (paper Table 2); accuracy — not comparable to AUROC rows; PaPaGei 0.80 / Chronos 0.77 / NormWear 0.77","open_weights":false}]},{"id":"mwl__wesad__exact_match","dataset":"wesad","dataset_label":"WESAD","metric":"exact_match","metric_label":null,"higher_is_better":true,"rows":[{"rank":1,"method":"PulseLM (Qwen2.5-7B + Pulse-PPG)","method_short":"PulseLM","entry_slug":"pulselm-ppg-text-qa-2026","entry_kind":"paper","value":0.497,"value_text":"0.497","protocol":"10 s @125 Hz, subject-disjoint ~8/1/1 per source; categorical QA exact-match (arXiv 2603.03331 Table 2); stress label set appears multi-class (Blind 0.183); not comparable to binary acc/AUROC rows","open_weights":"HF checkpoint linked; code CC BY 4.0"}]}]},{"task":"sleep-staging","task_name":"Sleep staging","total_rows":5,"boards":[{"id":"sleep-staging__dreamt__auroc","dataset":"dreamt","dataset_label":"DREAMT","metric":"auroc","metric_label":"Area under ROC","higher_is_better":true,"rows":[{"rank":1,"method":"Multimodal Self-Supervised Learning for Wearable Sleep Staging Using Photoplethysmography and Accelerometer Signals","method_short":"Multimodal Self-Supervised Learning for Wearable","entry_slug":"lee-multimodal-ssl-ppg-acc-sleep-icassp-2026","entry_kind":"paper","value":0.72,"value_text":"0.72","protocol":"DREAMT 4-class staging; SSL PPG+ACC pretrain (23,000 h unlabeled) + causal Mamba head; vs supervised 0.59; abstract-only, split unknown","open_weights":false}]},{"id":"sleep-staging__dreamt__cohen_kappa","dataset":"dreamt","dataset_label":"DREAMT","metric":"cohen_kappa","metric_label":"Cohen’s κ (e.g. sleep staging)","higher_is_better":true,"rows":[{"rank":1,"method":"Multimodal Self-Supervised Learning for Wearable Sleep Staging Using Photoplethysmography and Accelerometer Signals","method_short":"Multimodal Self-Supervised Learning for Wearable","entry_slug":"lee-multimodal-ssl-ppg-acc-sleep-icassp-2026","entry_kind":"paper","value":0.3,"value_text":"0.3","protocol":"DREAMT 4-class; vs supervised 0.06; abstract-only","open_weights":false}]},{"id":"sleep-staging__hmc__balanced_accuracy","dataset":"hmc","dataset_label":"hmc","metric":"balanced_accuracy","metric_label":"Balanced accuracy (HAR / class-imbalance)","higher_is_better":true,"rows":[{"rank":1,"method":"SpecBraM: What Should an EEG Foundation Model Predict? Masked Band-Power Prediction versus Waveform Reconstruction","method_short":"SpecBraM","entry_slug":"specbram-masked-band-power-eeg-fm-2026","entry_kind":"paper","value":0.7425,"value_text":"0.7425","protocol":"HMC sleep staging, EEG, frozen features + strict linear probe, mean ± sd over 3 pretrain seeds (±0.0045); full fine-tune 0.7669","open_weights":false}]},{"id":"sleep-staging__isruc-s1__balanced_accuracy","dataset":"isruc-s1","dataset_label":"isruc-s1","metric":"balanced_accuracy","metric_label":"Balanced accuracy (HAR / class-imbalance)","higher_is_better":true,"rows":[{"rank":1,"method":"SpecBraM: What Should an EEG Foundation Model Predict? Masked Band-Power Prediction versus Waveform Reconstruction","method_short":"SpecBraM","entry_slug":"specbram-masked-band-power-eeg-fm-2026","entry_kind":"paper","value":0.7916,"value_text":"0.7916","protocol":"ISRUC subgroup I, EEG, frozen features + strict linear probe, mean ± sd over 3 pretrain seeds (±0.0081); full fine-tune 0.8107","open_weights":false}]},{"id":"sleep-staging__opentslm-sleep-cot__macro_f1","dataset":"opentslm-sleep-cot","dataset_label":"opentslm-sleep-cot","metric":"macro_f1","metric_label":"Macro-averaged F1","higher_is_better":true,"rows":[{"rank":1,"method":"lapras-latent-reasoning-time-series-language-models-2026 (Lapras on ChatTS-7B)","method_short":"Lapras","entry_slug":"lapras-latent-reasoning-time-series-language-models-2026","entry_kind":"paper","value":79.49,"value_text":"79.49 (%)","protocol":"Multiple-choice sleep-stage QA (OpenTSLM, Langer et al. 2026), 923 test items, 1-ch 1500-step windows; same-backbone No-CoT 78.05, Explicit-CoT 65.48.","open_weights":null}]}]},{"task":"potassium-estimation","task_name":"Potassium estimation","total_rows":1,"boards":[{"id":"potassium-estimation__mc-med__pearson_r","dataset":"mc-med","dataset_label":"mc-med","metric":"pearson_r","metric_label":"Pearson correlation (labs / continuous)","higher_is_better":true,"rows":[{"rank":1,"method":"AnyPPG","method_short":"AnyPPG","entry_slug":"ppg-fm-decodability-audit-clinical-claims-iclr2027-sub","entry_kind":"paper","value":0.21,"value_text":"0.21","protocol":"model=AnyPPG; frozen encoder probe, 6,348 encounter-matched ED encounters; falls to 0.05 after age + renal adjustment","open_weights":true}]}]}]}