Semi-supervised classification on SVHN 1000 labels
16.61Error RateADGM
Evaluation Results
| Method | Links | |
|---|---|---|
| ADGM2015.11 | 16.61 | |
| Adversarial Autoencoderstraining=end-to-end2015.11 | 17.7 | |
| ADGMlabeled samples=10002017.06 | 22.86 | |
| VAT2015.11 | 24.63 | |
| VATlabeled samples=10002017.06 | 24.63 | |
| SHOT-VAEBackbone=4-layer MLP, Latent dimension (z)=32, Number of labels=1000, Number of parameters=~1M2020.11 | 28.82 | |
| smooth-ELBOBackbone=4-layer MLP, Latent dimension (z)=32, Number of labels=1000, Number of parameters=~1M2020.11 | 29.38 | |
| Domain-VAEBackbone=4-layer MLP, Latent dimension (z)=32, Number of labels=1000, Number of parameters=~1M2020.11 | 32.17 | |
| VAE (M1 + M2)2015.11 | 36.02 | |
| M1+M2Backbone=4-layer MLP, Latent dimension (z)=32, Number of labels=1000, Number of parameters=~1M2020.11 | 36.02 | |
| Disentangled-VAEBackbone=4-layer MLP, Latent dimension (z)=32, Number of labels=1000, Number of parameters=~1M2020.11 | 38.91 | |
| NN Baseline2015.11 | 47.5 | |
| M2Backbone=4-layer MLP, Latent dimension (z)=32, Number of labels=1000, Number of parameters=~1M2020.11 | 54.33 | |
| VAE (M1) + TSVM2015.11 | 55.33 |