Image Classification on CIFAR-100 (test) (Accuracy, Calibration, and Detection Metrics)
82.5AccuracyDeep Ensembles
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Deep EnsemblesBackbone=WideResNet-28×10, ΔParam=109.64, Time (h)=16.6, M=42023.12 | 82.5 | 0.903 | 22.9 | 81.6 | 67.9 | 71.3 | |
| BatchEnsembleBackbone=WideResNet-28×10, ΔParam=0.1, Time (h)=25.6, M=42023.12 | 82.3 | 0.835 | 13 | 88.1 | 78.2 | 69.8 | |
| Deep EnsemblesBackbone=ResNet-50, ΔParam=71.12, Time (h)=6.8, M=42023.12 | 80.9 | 0.713 | 2.6 | 89.2 | 80.8 | 52.5 | |
| ABNNBackbone=WideResNet-28×10, ΔParam=0.05, Time (h)=5, M=42023.12 | 80.4 | 1.08 | 5.5 | 85 | 75 | 57.7 | |
| Single ModelBackbone=WideResNet-28×10, Time (h)=4.22023.12 | 80.3 | 0.963 | 15.6 | 81 | 64.2 | 80.1 | |
| MIMO (ρ = 1)Backbone=WideResNet-28×10, ΔParam=0.19, Time (h)=12.6, M=42023.12 | 80.2 | 0.822 | 2.8 | 84.9 | 72 | 72.8 | |
| LaplaceBackbone=WideResNet-28×10, Time (h)=4.2, M=42023.12 | 80.1 | 0.942 | 16 | 83.4 | 72.1 | 59.9 | |
| LPBNNBackbone=WideResNet-28×10, ΔParam=0.72, Time (h)=23.3, M=42023.12 | 79.7 | 0.831 | 7 | 79 | 70.1 | 71.4 | |
| MIMO (ρ = 1)Backbone=ResNet-50, ΔParam=0.63, Time (h)=6.7, M=42023.12 | 79 | 0.876 | 7.9 | 87.5 | 76.9 | 64.7 | |
| ABNNBackbone=ResNet-50, ΔParam=0.16, Time (h)=2, M=42023.12 | 78.9 | 0.889 | 5.5 | 89.4 | 81 | 50.1 | |
| LPBNNBackbone=ResNet-50, ΔParam=1.83, Time (h)=17.2, M=42023.12 | 78.5 | 1.02 | 11.3 | 88.2 | 77.8 | 73.5 | |
| Single ModelBackbone=ResNet-50, Time (h)=1.72023.12 | 78.3 | 0.905 | 8.9 | 87.4 | 77.9 | 57.6 | |
| LaplaceBackbone=ResNet-50, Time (h)=1.7, M=42023.12 | 78.2 | 0.987 | 14.2 | 89.2 | 81 | 51.8 | |
| BatchEnsembleBackbone=ResNet-50, ΔParam=0.11, Time (h)=17.2, M=42023.12 | 66.6 | 1.788 | 18.2 | 85.2 | 74.6 | 60.6 |