Image Classification on CIFAR-10 (test) (Accuracy, NLL, ECE, and Rank Metrics)
94.4AccuracyMAP
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MAPBackbone=ResNet-56, Monte-Carlo samples=642026.06 | 94.4 | 0.252 | 3.7 | — | |
| MF-VIBackbone=ResNet-56, Monte-Carlo samples=642026.06 | 94.4 | 0.188 | 1.6 | — | |
| GP - SubsetBackbone=ResNet-56, Monte-Carlo samples=642026.06 | 94.4 | 0.403 | 22.1 | — | |
| LLA*Backbone=ResNet-56, Monte-Carlo samples=642026.06 | 94.4 | 0.213 | 2.2 | — | |
| LLA* KFACBackbone=ResNet-56, Monte-Carlo samples=642026.06 | 94.4 | 0.202 | 2.4 | — | |
| ELLABackbone=ResNet-56, Monte-Carlo samples=64, M=2000, K=202026.06 | 94.4 | 0.187 | 0.7 | 2.37 | |
| Sampled LLABackbone=ResNet-56, Monte-Carlo samples=642026.06 | 94.4 | 0.185 | 1.5 | 2 | |
| VaLLABackbone=ResNet-56, Monte-Carlo samples=642026.06 | 94.4 | 0.183 | 0.9 | 1.37 | |
| MAPBackbone=ResNet-44, Monte-Carlo samples=642026.06 | 94 | 0.275 | 3.9 | — | |
| LLA*Backbone=ResNet-44, Monte-Carlo samples=642026.06 | 94 | 0.237 | 2.8 | — | |
| LLA* KFACBackbone=ResNet-44, Monte-Carlo samples=642026.06 | 94 | 0.232 | 2.8 | — | |
| Sampled LLABackbone=ResNet-44, Monte-Carlo samples=642026.06 | 94 | 0.2 | 0.7 | 2 | |
| VaLLABackbone=ResNet-44, Monte-Carlo samples=642026.06 | 94 | 0.198 | 0.8 | 1.37 | |
| MF-VIBackbone=ResNet-44, Monte-Carlo samples=642026.06 | 93.9 | 0.206 | 1.8 | — | |
| ELLABackbone=ResNet-44, Monte-Carlo samples=64, M=2000, K=202026.06 | 93.9 | 0.204 | 0.7 | 2.37 | |
| SNGPBackbone=ResNet-44, Monte-Carlo samples=642026.06 | 93.8 | 0.242 | 2.8 | — | |
| SNGPBackbone=ResNet-56, Monte-Carlo samples=642026.06 | 93.8 | 0.229 | 2.2 | — | |
| GP - SubsetBackbone=ResNet-44, Monte-Carlo samples=642026.06 | 93.6 | 0.424 | 22.5 | — | |
| MAPBackbone=ResNet-32, Monte-Carlo samples=642026.06 | 93.5 | 0.292 | 4.1 | — | |
| MF-VIBackbone=ResNet-32, Monte-Carlo samples=642026.06 | 93.5 | 0.222 | 2 | — | |
| LLA*Backbone=ResNet-32, Monte-Carlo samples=642026.06 | 93.5 | 0.259 | 3.3 | — | |
| LLA* KFACBackbone=ResNet-32, Monte-Carlo samples=642026.06 | 93.5 | 0.26 | 3.3 | — | |
| ELLABackbone=ResNet-32, Monte-Carlo samples=64, M=2000, K=202026.06 | 93.5 | 0.215 | 0.8 | 2.37 | |
| Sampled LLABackbone=ResNet-32, Monte-Carlo samples=642026.06 | 93.5 | 0.217 | 0.8 | 2 | |
| VaLLABackbone=ResNet-32, Monte-Carlo samples=642026.06 | 93.5 | 0.211 | 0.7 | 1.37 | |
| GP - SubsetBackbone=ResNet-32, Monte-Carlo samples=642026.06 | 93.4 | 0.462 | 24.7 | — | |
| SNGPBackbone=ResNet-32, Monte-Carlo samples=642026.06 | 93.2 | 0.256 | 2.5 | — | |
| LLA DiagBackbone=ResNet-56, Monte-Carlo samples=642026.06 | 92.9 | 0.843 | 48 | — | |
| LLA DiagBackbone=ResNet-44, Monte-Carlo samples=642026.06 | 92.8 | 0.778 | 44.5 | — | |
| MF-VIBackbone=ResNet-20, Monte-Carlo samples=642026.06 | 92.7 | 0.231 | 1.6 | — | |
| LLA DiagBackbone=ResNet-32, Monte-Carlo samples=642026.06 | 92.7 | 0.755 | 43 | — | |
| MAPBackbone=ResNet-20, Monte-Carlo samples=642026.06 | 92.6 | 0.282 | 3.9 | — | |
| GP - SubsetBackbone=ResNet-20, Monte-Carlo samples=642026.06 | 92.6 | 0.555 | 29.9 | — | |
| LLA*Backbone=ResNet-20, Monte-Carlo samples=642026.06 | 92.6 | 0.269 | 3.4 | — | |
| LLA* KFACBackbone=ResNet-20, Monte-Carlo samples=642026.06 | 92.6 | 0.271 | 3.5 | — | |
| VaLLABackbone=ResNet-20, Monte-Carlo samples=642026.06 | 92.6 | 0.228 | 0.7 | 1.37 | |
| ELLABackbone=ResNet-20, Monte-Carlo samples=64, M=2000, K=202026.06 | 92.5 | 0.233 | 0.9 | 2.37 | |
| Sampled LLABackbone=ResNet-20, Monte-Carlo samples=642026.06 | 92.5 | 0.231 | 0.6 | 2 | |
| SNGPBackbone=ResNet-20, Monte-Carlo samples=642026.06 | 92.4 | 0.266 | 2.4 | — | |
| LLA DiagBackbone=ResNet-20, Monte-Carlo samples=642026.06 | 92.2 | 0.728 | 40.4 | — | |
| LLA KFACBackbone=ResNet-20, Monte-Carlo samples=642026.06 | 92 | 0.852 | 46.7 | — | |
| LLA KFACBackbone=ResNet-32, Monte-Carlo samples=642026.06 | 91.8 | 1.027 | 54.7 | — | |
| LLA KFACBackbone=ResNet-44, Monte-Carlo samples=642026.06 | 91.4 | 1.091 | 56.6 | — | |
| LLA KFACBackbone=ResNet-56, Monte-Carlo samples=642026.06 | 89.8 | 1.174 | 57.9 | — |