Table 15 Classification accuracy comparison with various SSL methods.
From: Deep representation learning using layer-wise VICReg losses
SSL method/Dataset | MNIST (%) | Fashion MNIST (%) |
|---|---|---|
SimCLR | 97.2 | 74.9 |
MoCo | 78.6 | 65.1 |
MoCov2 | 94.6 | 71.2 |
VICReg | 96.7 | 73.2 |
SimVAE | 98.4 | 82.1 |
Ours (fine-tuned with 500 samples) | 85.13 | 77.49 |
Ours (fine-tuned with 1000 samples) | 88.66 | 79.88 |
Ours (fine-tuned with 2500 samples) | 91.69 | 82.57 |
Ours (fine-tuned with 5000 samples) | 94.23 | 83.65 |
