Adversarial Robustness on CIFAR-10 (PGD/C&W Norm Evaluation)
49.71PGD Accuracy (L-inf)Muon
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MuonTraining Norm=l_inf, Optimizer=Muon, Checkpoint=Best, Backbone=WRN-34-202026.05 | 49.71 | 48.56 | 60.39 | 59.07 | |
| SGDNorm=l_inf, Checkpoint=Best, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 48.07 | 48.11 | 60 | 59.19 | |
| MuonTraining Norm=l_inf, Optimizer=Muon, Checkpoint=Last, Backbone=WRN-34-202026.05 | 47.71 | 48.01 | 57.83 | 58.08 | |
| MuonNorm=l_inf, Checkpoint=Best, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 47.03 | 45.66 | 61.15 | 59.19 | |
| SGDTraining Norm=l_inf, Optimizer=SGD, Checkpoint=Best, Backbone=WRN-34-202026.05 | 45.13 | 43.08 | 57.92 | 55.81 | |
| MuonNorm=l_inf, Checkpoint=Last, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 45.12 | 45.78 | 62.07 | 61.97 | |
| AdamWTraining Norm=l_inf, Optimizer=AdamW, Checkpoint=Best, Backbone=WRN-34-202026.05 | 44.81 | 45.05 | 50.96 | 50.22 | |
| AdamWTraining Norm=l_inf, Optimizer=AdamW, Checkpoint=Last, Backbone=WRN-34-202026.05 | 44.17 | 43.17 | 50.6 | 49.25 | |
| SGDNorm=l_inf, Checkpoint=Last, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 43.78 | 44.19 | 58.54 | 58.44 | |
| SGDNorm=l_inf + l_1, Checkpoint=Best, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 43.44 | 43.25 | 66.29 | 65.32 | |
| MuonTraining Norm=l_inf + l_1, Optimizer=Muon, Checkpoint=Best, Backbone=WRN-34-202026.05 | 43.03 | 42.41 | 63.19 | 61.66 | |
| MuonNorm=l_inf + l_1, Checkpoint=Best, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 42.76 | 42.1 | 65.25 | 63.56 | |
| AdamWNorm=l_1, Checkpoint=Last, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 42.27 | 40.65 | 56.04 | 55.82 | |
| MuonTraining Norm=l_inf + l_1, Optimizer=Muon, Checkpoint=Last, Backbone=WRN-34-202026.05 | 41.78 | 42.07 | 62.82 | 62.96 | |
| AdamWNorm=l_inf, Checkpoint=Best, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 41.76 | 40.95 | 49.55 | 48.53 | |
| SGDTraining Norm=l_inf + l_1, Optimizer=SGD, Checkpoint=Best, Backbone=WRN-34-202026.05 | 40.58 | 39.12 | 62.14 | 60.47 | |
| SGDTraining Norm=l_inf, Optimizer=SGD, Checkpoint=Last, Backbone=WRN-34-202026.05 | 40.13 | 40.47 | 54.9 | 55.36 | |
| MuonNorm=l_inf + l_1, Checkpoint=Last, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 39.97 | 40.89 | 68.09 | 67.78 | |
| AdamWNorm=l_inf + l_1, Checkpoint=Best, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 39.24 | 39.55 | 56.56 | 55.67 | |
| AdamWTraining Norm=l_inf + l_1, Optimizer=AdamW, Checkpoint=Best, Backbone=WRN-34-202026.05 | 38.34 | 37.86 | 59.5 | 58.12 | |
| AdamWTraining Norm=l_inf + l_1, Optimizer=AdamW, Checkpoint=Last, Backbone=WRN-34-202026.05 | 38.34 | 37.86 | 59.49 | 58.12 | |
| SGDNorm=l_inf + l_1, Checkpoint=Last, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 37.3 | 38.23 | 64.96 | 64.67 | |
| AdamWNorm=l_inf, Checkpoint=Last, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 36.7 | 35.48 | 51.64 | 50.15 | |
| SGDTraining Norm=l_inf + l_1, Optimizer=SGD, Checkpoint=Last, Backbone=WRN-34-202026.05 | 35.23 | 35.53 | 62.25 | 62.42 | |
| MuonTraining Norm=l_2, Optimizer=Muon, Checkpoint=Best, Backbone=WRN-34-202026.05 | 30.71 | 30.88 | 70.37 | 70.49 | |
| MuonTraining Norm=l_2, Optimizer=Muon, Checkpoint=Last, Backbone=WRN-34-202026.05 | 30.29 | 30.5 | 70.55 | 70.69 | |
| MuonNorm=l_2, Checkpoint=Best, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 29.88 | 31.71 | 65.23 | 64.75 | |
| AdamWTraining Norm=l_2, Optimizer=AdamW, Checkpoint=Last, Backbone=WRN-34-202026.05 | 27.53 | 26.52 | 61.33 | 62.56 | |
| SGDNorm=l_2, Checkpoint=Best, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 27.15 | 28.94 | 65.95 | 66.32 | |
| AdamWTraining Norm=l_2, Optimizer=AdamW, Checkpoint=Best, Backbone=WRN-34-202026.05 | 26.83 | 26.58 | 61.62 | 62.95 | |
| SGDTraining Norm=l_2, Optimizer=SGD, Checkpoint=Best, Backbone=WRN-34-202026.05 | 26.23 | 26.38 | 64.42 | 64.66 | |
| AdamWNorm=l_2, Checkpoint=Best, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 25.99 | 28.48 | 59.71 | 59.97 | |
| MuonNorm=l_1, Checkpoint=Best, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 25.97 | 27.03 | 59.39 | 58.32 | |
| SGDTraining Norm=l_2, Optimizer=SGD, Checkpoint=Last, Backbone=WRN-34-202026.05 | 25.67 | 25.78 | 65.06 | 65.22 | |
| MuonNorm=l_2, Checkpoint=Last, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 25.12 | 25.8 | 67.06 | 67.4 | |
| AdamWNorm=l_2, Checkpoint=Last, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 24.84 | 27.62 | 59.82 | 60.38 | |
| AdamWNorm=l_inf + l_1, Checkpoint=Last, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 24.74 | 24.89 | 57.87 | 56.12 | |
| AdamWNorm=l_1, Checkpoint=Best, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 24.28 | 26.39 | 55.62 | 55.19 | |
| MuonTraining Norm=l_1, Optimizer=Muon, Checkpoint=Best, Backbone=WRN-34-202026.05 | 24.2 | 24.43 | 65.88 | 66.05 | |
| SGDTraining Norm=l_1, Optimizer=SGD, Checkpoint=Best, Backbone=WRN-34-202026.05 | 23.91 | 23.8 | 43.44 | 42.51 | |
| AdamWTraining Norm=l_1, Optimizer=AdamW, Checkpoint=Best, Backbone=WRN-34-202026.05 | 23.79 | 26.23 | 53.03 | 52.74 | |
| MuonTraining Norm=l_1, Optimizer=Muon, Checkpoint=Last, Backbone=WRN-34-202026.05 | 23.65 | 23.74 | 65.36 | 65.44 | |
| SGDNorm=l_2, Checkpoint=Last, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 21.27 | 23.46 | 63.58 | 64.02 | |
| AdamWTraining Norm=l_1, Optimizer=AdamW, Checkpoint=Last, Backbone=WRN-34-202026.05 | 21.26 | 22.64 | 57.92 | 59.11 | |
| SGDTraining Norm=l_1, Optimizer=SGD, Checkpoint=Last, Backbone=WRN-34-202026.05 | 20.04 | 20.2 | 60.65 | 60.79 | |
| MuonNorm=l_1, Checkpoint=Last, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 19.76 | 20.81 | 63.67 | 64.23 | |
| SGDNorm=l_1, Checkpoint=Last, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 19.11 | 19.62 | 59.53 | 60.01 | |
| SGDNorm=l_1, Checkpoint=Best, Backbone=PreActResNet-18, epsilon_inf=8/255, epsilon_2=0.52026.05 | 19.02 | 19.51 | 59.44 | 59.95 |