Table 5 Comparison results of different backbones (ResNet18, ResNet34, ResNet50, and MSDNN) on CPSC (ECG) dataset

From: Transforming label-efficient decoding of healthcare wearables with self-supervised learning and “embedded” medical domain expertise

Methods

ResNet18

ResNet34

ResNet50

MSDNN

SimCLR9

0.532

0.554

0.574

0.464

BYOL37

0.428

0.400

0.527

0.328

MoCo38

0.524

0.486

0.568

0.493

NNCLR39

0.517

0.515

0.555

0.424

TS40

0.549

0.559

0.562

0.505

SwAV41

0.510

0.520

0.518

0.480

AMCL42

0.503

0.537

0.563

0.501

CLOCS10

0.537

0.557

0.560

0.533

TFC15

0.411

0.412

0.550

0.373

SoftIns43

0.557

0.532

0.565

0.508

RNC49

0.550

0.542

0.563

0.547

Ours

0.574

0.580

0.581

0.555

  1. We compared the class-average-average F1 on the test subset, based on the model with the best F1 on the validation subset. Bold indicates the best result.
Back to article page