Image Classification on CIFAR-10 (Accuracy, Delta, Correlation)
93.92AccuracyPolar
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PolarFLOPs Reduction (%)=54, Epochs=2002021.07 | 93.92 | 0.04 | — | |
| NSFLOPs Reduction (%)=51, Epochs=2002021.07 | 93.62 | -0.26 | — | |
| IndependentFLOPs Reduction (%)=0, Epochs=1602021.07 | 93.6 | — | 93.02 | |
| MagFLOPs Reduction (%)=50, Epochs=2002021.07 | 93.56 | -0.04 | 93.15 | |
| ID+LRankFLOPs Reduction (%)=50, Epochs=02021.07 | 93.43 | 0.31 | 95.9 | |
| HRankFLOPs Reduction (%)=54, Epochs=4802021.07 | 93.43 | -0.5 | — | |
| IDFLOPs Reduction (%)=50, Epochs=602021.07 | 93.31 | 0.3 | 97.29 | |
| UFOComm. (GB)=0.366, Lat. (s)=13.82, Backbone=ResNet-202026.02 | 92.57 | — | — | |
| WinoVidiViciComm. (GB)=0.964, Lat. (s)=60.4, Backbone=ResNet-202026.02 | 92.29 | — | — | |
| Ours (IE-CL)Backbone=ResNet-18, Batch Size=256, Evaluation Protocol=Linear probe2026.03 | 92.1 | — | — | |
| BYOLBackbone=ResNet-18, Batch Size=256, Evaluation Protocol=Linear probe2026.03 | 91.9 | — | — | |
| LegendreComm. (GB)=0.965, Lat. (s)=61.86, Backbone=ResNet-202026.02 | 91.8 | — | — | |
| MoCo-v3Backbone=ResNet-18, Batch Size=256, Evaluation Protocol=Linear probe2026.03 | 91.8 | — | — | |
| MoCo-v2Backbone=ResNet-18, Batch Size=256, Evaluation Protocol=Linear probe2026.03 | 91.3 | — | — | |
| SimSiamBackbone=ResNet-18, Batch Size=256, Evaluation Protocol=Linear probe2026.03 | 91.2 | — | — | |
| SimCLRBackbone=ResNet-18, Batch Size=256, Evaluation Protocol=Linear probe2026.03 | 91.1 | — | — | |
| S3OCBackbone=ResNet-18, Batch Size=256, Evaluation Protocol=Linear probe2026.03 | 91 | — | — | |
| MinEntBackbone=ResNet-18, Batch Size=256, Evaluation Protocol=Linear probe2026.03 | 90.8 | — | — | |
| BQWComm. (GB)=0.783, Lat. (s)=56.16, Backbone=ResNet-202026.02 | 90.74 | — | — | |
| W-MSEBackbone=ResNet-18, Batch Size=256, Evaluation Protocol=Linear probe2026.03 | 90.6 | — | — | |
| PAWComm. (GB)=0.784, Lat. (s)=57.01, Backbone=ResNet-202026.02 | 90.25 | — | — | |
| MGD3+EVLFBackbone=ConvNet-4, Pre-training Dataset=distilled ImageNet-1K subset (0.8x), Evaluation Protocol=Fine-tuning2026.03 | 90.23 | — | — | |
| KRR-STBackbone=ConvNet-4, Pre-training Dataset=distilled ImageNet-1K subset (0.8x), Evaluation Protocol=Fine-tuning2026.03 | 89.33 | — | — | |
| w/o preBackbone=ConvNet-4, Evaluation Protocol=Fine-tuning2026.03 | 88.66 | — | — | |
| RandomBackbone=ConvNet-4, Evaluation Protocol=Fine-tuning2026.03 | 88.46 | — | — | |
| FRePoBackbone=ConvNet-4, Pre-training Dataset=distilled ImageNet-1K subset (0.8x), Evaluation Protocol=Fine-tuning2026.03 | 87.88 | — | — | |
| DeepClusterBackbone=ResNet-18, Batch Size=256, Evaluation Protocol=Linear probe2026.03 | 84.3 | — | — | |
| PTQComm. (GB)=0.783, Lat. (s)=56.85, Backbone=ResNet-202026.02 | 80.63 | — | — |