Image Classification on MNIST Aff(2, R) (test)
96.95AccuracyResNet18 + TIED
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ResNet18 + TIEDEnergy type=VAE evidence lower bound (+ adv. reg.)2026.02 | 96.95 | 0.71 | |
| affConv / homConv2026.02 | 95.08 | — | |
| ResNet18 + LieLACEnergy type=VAE evidence lower bound (+ adv. reg.)2026.02 | 94.38 | 0.91 | |
| ResNet18 + FoCalEnergy type=VAE evidence lower bound (+ adv. reg.)2026.02 | 86.38 | 3.39 | |
| ResNet18 + TIEDEnergy type=Classifier logit confidence2026.02 | 85.64 | 5.05 | |
| ResNet18 + Kinetic LangevinEnergy type=VAE evidence lower bound (+ adv. reg.)2026.02 | 74.72 | 7.15 | |
| ResNet18 + LieLACEnergy type=Classifier logit confidence2026.02 | 73.79 | 7.03 | |
| ResNet18 + ITSEnergy type=Classifier logit confidence2026.02 | 67.18 | 9.9 | |
| ResNet18 + FoCalEnergy type=Classifier logit confidence2026.02 | 66.97 | 10.92 | |
| ResNet182026.02 | 55.48 | — | |
| ResNet18 + Kinetic LangevinEnergy type=Classifier logit confidence2026.02 | 53.98 | 15.29 | |
| ResNet18 + ITSEnergy type=VAE evidence lower bound (+ adv. reg.)2026.02 | 45.97 | 10.89 |