Image Classification on CIFAR-100 WRN-28-10 (test)
85.59Top-1 AccuracyCut-MixMo + CutMix
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Cut-MixMo + CutMixBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=4/12021.03 | 85.59 | 97.33 | 53.2 | 53.3 | 1.95 | |
| DE (2 Nets) + CutMixBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=2/22021.03 | 85.46 | 96.9 | 57.4 | 57.5 | 1.89 | |
| Cut-MixMoBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=4/12021.03 | 85.32 | 97.12 | 53.6 | 54.8 | 4.53 | |
| Mixup BABackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=7/12021.03 | 85.25 | 96.85 | — | — | — | |
| Linear-MixMo + CutMixBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=4/12021.03 | 85.24 | 96.97 | 56.3 | 56.4 | 3.53 | |
| Cut-MixMo + CutMixBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=2/12021.03 | 85.17 | 97.28 | 54.4 | 54.5 | 2.13 | |
| Linear-MixMo + CutMixBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=2/12021.03 | 84.55 | 96.95 | 57.4 | 57.5 | 2.54 | |
| DE (2 Nets)Backbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=2/22021.03 | 84.3 | 96.35 | — | — | — | |
| Cut-MixMoBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=2/12021.03 | 84.07 | 96.97 | 56.6 | 57.9 | 4.19 | |
| GradAug + CutMixBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=2/12021.03 | 84.05 | 96.08 | 65.4 | 66.9 | 2.76 | |
| Puzzle-MixBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=3/12021.03 | 83.98 | 96.28 | 68.1 | — | — | |
| Hard PatchUpBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=1/12021.03 | 83.87 | 96 | 66 | — | — | |
| CutMixBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=1/12021.03 | 83.74 | 96.18 | 66.1 | — | — | |
| MixupBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=1/12021.03 | 83.15 | 95.75 | 66.3 | 67.3 | 1.62 | |
| MIMOBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=4/1, M=22021.03 | 82.74 | 95.9 | 67 | 74 | 7.56 | |
| Linear-MixMoBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=4/12021.03 | 82.53 | 96.08 | 65.8 | 68.5 | 6.64 | |
| MIMOBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=2/1, M=22021.03 | 82.04 | 95.75 | 69.1 | 72.4 | 3.62 | |
| MIMOBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=4/1, M=32021.03 | 82 | — | 69 | — | 2.2 | |
| Linear-MixMoBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=2/12021.03 | 81.88 | 95.97 | 67.8 | 70.3 | 6.2 | |
| VanillaBackbone=WRN-28-10, Early stopping=false, Time Tr./Inf.=1/12021.03 | 81.47 | 95.57 | 73.6 | 76.2 | 6.47 | |
| System-aware Biased Local SGDEpochs=600, Comm. per epoch=4, Wall-clock Time (sec)=7556.48, Hardware=2-CPU + 8-GPU, Backbone=WRN282025.08 | 80.58 | — | — | — | — | |
| Synchronous SGD + Data-ParallelismEpochs=200, Comm. per epoch=160, Wall-clock Time (sec)=48180.65, Hardware=8-GPU only, Backbone=WRN282025.08 | 80.26 | — | — | — | — | |
| Synchronous SGD + Data-ParallelismEpochs=200, Comm. per epoch=128, Wall-clock Time (sec)=80602.47, Hardware=2-CPU + 8-GPU, Backbone=WRN282025.08 | 80.2 | — | — | — | — | |
| BytePSEpochs=200, Comm. per epoch=160, Wall-clock Time (sec)=130155.6, Hardware=2-CPU + 8-GPU, Backbone=WRN282025.08 | 80.07 | — | — | — | — | |
| System-aware Biased Local SGDEpochs=400, Comm. per epoch=4, Wall-clock Time (sec)=5037.66, Hardware=2-CPU + 8-GPU, Backbone=WRN282025.08 | 80.03 | — | — | — | — | |
| Balanced Local SGDEpochs=200, Comm. per epoch=4, Wall-clock Time (sec)=43952.17, Hardware=2-CPU + 8-GPU, Backbone=WRN282025.08 | 78.24 | — | — | — | — | |
| System-aware Biased Local SGDEpochs=200, Comm. per epoch=4, Wall-clock Time (sec)=2518.83, Hardware=2-CPU + 8-GPU, Backbone=WRN282025.08 | 77.97 | — | — | — | — | |
| DLBEpochs=200, Comm. per epoch=128, Wall-clock Time (sec)=38544.52, Hardware=2-CPU + 8-GPU, Backbone=WRN282025.08 | 76.42 | — | — | — | — |