Generative Modeling on MNIST binarized (test)
78.43NLLλ-VAE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| λ-VAEdelta=1.01, Architecture=three-layer MLP, K=302026.07 | 78.43 | 0.144 | 0.9667 | 69.4 | |
| Free BitsArchitecture=Deeper convolutional, K=302026.07 | 79.1 | 0.145 | — | — | |
| AVBArchitecture=Deeper convolutional, K=302026.07 | 80.24 | 0.147 | — | — | |
| InfoVAEArchitecture=Deeper convolutional, K=302026.07 | 80.76 | 0.148 | — | — | |
| VampPriorArchitecture=three-layer MLP, K=302026.07 | 85.57 | 0.157 | — | — | |
| λ-VAEdelta=1.19, Architecture=three-layer MLP, K=302026.07 | 93.8 | 0.172 | 0.7 | 29.5 | |
| λ-VAEdelta=1.10, Architecture=three-layer MLP, K=302026.07 | 94.44 | 0.173 | 0.7667 | 37.2 | |
| β-VAEbeta=2.0, Architecture=three-layer MLP, K=302026.07 | 96.73 | 0.178 | 0.3667 | 16.2 | |
| Standard VAElambda=1.0, Architecture=three-layer MLP, K=302026.07 | 97.17 | 0.178 | 0.4667 | 23.2 | |
| β-VAEbeta=0.5, Architecture=three-layer MLP, K=302026.07 | 102.94 | 0.189 | 0.6 | 34.2 |