Averaging Weights Leads to Wider Optima and Better Generalization

Open paper
Benchmark Model Rank Results
image-classification-on-cifar-10ShakeShake-2x64d + SWA#90Percentage correct: 97.12
image-classification-on-cifar-10WRN-28-10 + SWA#99Percentage correct: 96.79
image-classification-on-cifar-100PyramidNet-272 + SWA#75Percentage correct: 84.16
image-classification-on-cifar-100WRN+SWA#106Percentage correct: 82.15
image-classification-on-imagenetResNet-152 + SWA#754Top 1 Accuracy: 78.94%
image-classification-on-imagenetDenseNet-161 + SWA#785Top 1 Accuracy: 78.44%

Rank counts only results with a code link.