Image Classification on MNIST PGL(3, R) (test)
97.43AccuracyResNet18 + TIED
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ResNet18 + TIEDEnergy type=VAE evidence lower bound (+ adv. reg.)2026.02 | 97.43 | 0.57 | |
| ResNet18 + TIEDEnergy=VAE evidence lower bound (+ adv. reg.)2026.02 | 97.43 | — | |
| ResNet18 + LieLACEnergy type=VAE evidence lower bound (+ adv. reg.)2026.02 | 97.39 | 0.58 | |
| ResNet18 + LieLACEnergy=VAE evidence lower bound (+ adv. reg.)2026.02 | 97.39 | — | |
| affConv / homConv2026.02 | 95.71 | — | |
| affConv / homConv2026.02 | 95.71 | — | |
| ResNet18 + Kinetic LangevinEnergy type=VAE evidence lower bound (+ adv. reg.)2026.02 | 93.75 | 0.77 | |
| ResNet18 + Kinetic LangevinEnergy=VAE evidence lower bound (+ adv. reg.)2026.02 | 93.75 | — | |
| ResNet18 + TIEDEnergy type=Classifier logit confidence2026.02 | 89.84 | 1.83 | |
| ResNet18 + TIEDEnergy=Classifier logit confidence2026.02 | 89.84 | — | |
| ResNet18 + FoCalEnergy type=VAE evidence lower bound (+ adv. reg.)2026.02 | 89.69 | 2.2 | |
| ResNet18 + FoCalEnergy=VAE evidence lower bound (+ adv. reg.)2026.02 | 89.69 | — | |
| ResNet18 + Kinetic LangevinEnergy type=Classifier logit confidence2026.02 | 88.25 | 2.48 | |
| ResNet18 + Kinetic LangevinEnergy=Classifier logit confidence2026.02 | 88.25 | — | |
| ResNet182026.02 | 87.95 | — | |
| ResNet182026.02 | 87.95 | — | |
| ResNet18 + FoCalEnergy type=Classifier logit confidence2026.02 | 86.45 | 3.89 | |
| ResNet18 + FoCalEnergy=Classifier logit confidence2026.02 | 86.45 | — | |
| ResNet18 + LieLACEnergy type=Classifier logit confidence2026.02 | 85.93 | 2.66 | |
| ResNet18 + LieLACEnergy=Classifier logit confidence2026.02 | 85.93 | — |