Image Classification on CIFAR-10 Targeted 25% Noise (test)
84.04Test AccuracyDR-IS-online
Evaluation Results
| Method | Links | |
|---|---|---|
| DR-IS-onlineBackbone=VGG-19, Training Epochs=1602026.05 | 84.04 | |
| Uniform SGDBackbone=VGG-19, Training Epochs=1602026.05 | 74.52 | |
| RHO-LOSS (faithful)Backbone=VGG-19, Training Epochs=160, Proxy Backbone=ResNet-20, Holdout size=5k, Refresh Protocol=recomputed every 5 epochs2026.05 | 31.3 | |
| Standard ISBackbone=VGG-19, Training Epochs=1602026.05 | 10 | |
| RHO-LOSS (approx.)Backbone=VGG-19, Training Epochs=160, Proxy type=mean loss, Refresh=no refresh2026.05 | 10 |