Image Classification on ImageNet (val) (Training Time & Epochs)
201.6EpochsSOAP 0
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SOAP 0Optimization Algorithm=SOAP, Backbone=Deep Residual Network (ResNet), Convergence Threshold (Validation Loss)=< 1.3, Update Cadence=02025.02 | 201.6 | — | |
| EGOP up-frontOptimization Algorithm=AdamW, Backbone=Deep Residual Network (ResNet), Convergence Threshold (Validation Loss)=< 1.3, Reparameterization=up-front2025.02 | 150.98 | — | |
| Original CoordinatesOptimization Algorithm=AdamW, Backbone=Deep Residual Network (ResNet), Convergence Threshold (Validation Loss)=< 1.32025.02 | 24.49 | — | |
| SOAP 10000Optimization Algorithm=SOAP, Backbone=Deep Residual Network (ResNet), Convergence Threshold (Validation Loss)=< 1.3, Update Cadence=100002025.02 | 14.4 | — | |
| EGOP periodicOptimization Algorithm=AdamW, Backbone=Deep Residual Network (ResNet), Convergence Threshold (Validation Loss)=< 1.3, Reparameterization=periodic (once per epoch)2025.02 | 13.63 | — | |
| SOAP 100Optimization Algorithm=SOAP, Backbone=Deep Residual Network (ResNet), Convergence Threshold (Validation Loss)=< 1.3, Update Cadence=1002025.02 | 13.4 | — |