Image Classification on ColoredMNIST 1.0 (Din)
89.7AccuracyNaive-Independent
Evaluation Results
| Method | Links | |
|---|---|---|
| Naive-Independenttraining_process=standard VAE trained on unlabelled data, then classifier trained on latent projection2022.11 | 89.7 | |
| Naive-VAE-Classtype=standard VAE classifier2022.11 | 89 | |
| Chroma-VAElatent_representation=z1, classifier_type=MLP2022.11 | 89 | |
| Naive-Classtype=standard ERM classifier2022.11 | 87.8 | |
| Theoretical UB2022.11 | 75 | |
| Chroma-VAElatent_representation=z2, classifier_type=k-nearest neighbor (k=sqrt(N))2022.11 | 72.5 | |
| JTT2022.11 | 63.2 | |
| Invarianttraining_data=black-and-white images2022.11 | 60.8 |