Generative Modeling on MNIST (train)
136.73ELBOST
Evaluation results
| Method | Links | ||
|---|---|---|---|
| STCategorical × Latent Dimension=10×302026.03 | 136.73 | — | |
| STCategorical × Latent Dimension=8×42026.03 | 136.4 | — | |
| GST-1.0Categorical × Latent Dimension=8×42026.03 | 128.31 | — | |
| GumbelCategorical × Latent Dimension=8×42026.03 | 127.59 | — | |
| Gumbel-RaoCategorical × Latent Dimension=8×42026.03 | 125.65 | — | |
| ReinMax-RaoCategorical × Latent Dimension=8×42026.03 | 125.31 | — | |
| ReinMaxCategorical × Latent Dimension=8×42026.03 | 125.08 | — | |
| ReinMax-CVCategorical × Latent Dimension=8×42026.03 | 124.94 | — | |
| STCategorical × Latent Dimension=64×82026.03 | 113.97 | — | |
| STCategorical × Latent Dimension=8×162026.03 | 113.32 | — | |
| STCategorical × Latent Dimension=16×122026.03 | 113.31 | — | |
| Gumbel-RaoCategorical × Latent Dimension=64×82026.03 | 112.48 | — | |
| STCategorical × Latent Dimension=4×242026.03 | 112.21 | — | |
| Gumbel-RaoCategorical × Latent Dimension=16×122026.03 | 109.72 | — | |
| GumbelCategorical × Latent Dimension=64×82026.03 | 103.87 | — | |
| DisARMK=2, Architecture=2x200 VAE2023.04 | 102.75 | — | |
| GST-1.0Categorical × Latent Dimension=64×82026.03 | 102.41 | — | |
| Gumbel-RaoCategorical × Latent Dimension=10×302026.03 | 102.28 | — | |
| Double CVK=2, Architecture=2x200 VAE2023.04 | 102.14 | — | |
| RELAXK=3, Architecture=2x200 VAE2023.04 | 101.99 | — | |
| RODEOK=2, Architecture=2x200 VAE2023.04 | 101.89 | — | |
| GST-1.0Categorical × Latent Dimension=4×242026.03 | 101.64 | — | |
| GumbelCategorical × Latent Dimension=4×242026.03 | 101.49 | — | |
| ReinMaxCategorical × Latent Dimension=64×82026.03 | 101.09 | — | |
| Double CVK=3, Architecture=2x200 VAE2023.04 | 100.94 | — | |
| ARMSK=3, Architecture=2x200 VAE2023.04 | 100.84 | — | |
| ReinMax-CVCategorical × Latent Dimension=64×82026.03 | 100.66 | — | |
| RODEOK=3, Architecture=2x200 VAE2023.04 | 100.46 | — | |
| GumbelCategorical × Latent Dimension=16×122026.03 | 100.42 | — | |
| ReinMax-RaoCategorical × Latent Dimension=4×242026.03 | 100.41 | — | |
| ReinMax-RaoCategorical × Latent Dimension=64×82026.03 | 100.24 | — | |
| ReinMax-CVCategorical × Latent Dimension=4×242026.03 | 100.19 | — | |
| Gumbel-RaoCategorical × Latent Dimension=4×242026.03 | 100.14 | — | |
| ReinMaxCategorical × Latent Dimension=4×242026.03 | 99.88 | — | |
| GumbelCategorical × Latent Dimension=10×302026.03 | 99.8 | — | |
| Gumbel-RaoCategorical × Latent Dimension=8×162026.03 | 99.63 | — | |
| GumbelCategorical × Latent Dimension=8×162026.03 | 99.2 | — | |
| GST-1.0Categorical × Latent Dimension=10×302026.03 | 98.66 | — | |
| ReinMax-RaoCategorical × Latent Dimension=10×302026.03 | 98.61 | — | |
| ReinMaxCategorical × Latent Dimension=10×302026.03 | 98.5 | — | |
| GST-1.0Categorical × Latent Dimension=16×122026.03 | 98.37 | — | |
| GST-1.0Categorical × Latent Dimension=8×162026.03 | 98.31 | — | |
| ReinMax-CVCategorical × Latent Dimension=16×122026.03 | 98.21 | — | |
| ReinMaxK=2, Architecture=2x200 VAE2023.04 | 98.17 | — | |
| ReinMax-CVCategorical × Latent Dimension=10×302026.03 | 98.07 | — | |
| ReinMax-RaoCategorical × Latent Dimension=8×162026.03 | 98.03 | — | |
| ReinMaxCategorical × Latent Dimension=16×122026.03 | 97.98 | — | |
| ReinMaxK=3, Architecture=2x200 VAE2023.04 | 97.83 | — | |
| ReinMaxCategorical × Latent Dimension=8×162026.03 | 97.8 | — | |
| ReinMax-CVCategorical × Latent Dimension=8×162026.03 | 97.72 | — | |
| ReinMax-RaoCategorical × Latent Dimension=16×122026.03 | 97.62 | — | |
| DisARM-Treebatch size=200, training steps=5 x 10^5, latent dimensions=32, categorical dimensions=642023.04 | — | 103.1 | |
| GR-MCKbatch size=200, training steps=5 x 10^5, latent dimensions=32, categorical dimensions=642023.04 | — | 110.74 | |
| GR-MCKLatent dimensions (N x M)=8 x 42023.04 | — | 125.94 | |
| GR-MCKLatent dimensions (N x M)=4 x 242023.04 | — | 99.96 | |
| GR-MCKLatent dimensions (N x M)=8 x 162023.04 | — | 99.58 | |
| GR-MCKLatent dimensions (N x M)=16 x 122023.04 | — | 102.54 | |
| GR-MCKLatent dimensions (N x M)=64 x 82023.04 | — | 112.34 | |
| GR-MCKLatent dimensions (N x M)=10 x 302023.04 | — | 102.02 | |
| GR-MCKLatent dimensions (N x M)=AVG2023.04 | — | 107.06 | |
| GST-1.0batch size=200, training steps=5 x 10^5, latent dimensions=32, categorical dimensions=642023.04 | — | 96.09 | |
| GST-1.0Latent dimensions (N x M)=8 x 42023.04 | — | 126.35 | |
| GST-1.0Latent dimensions (N x M)=4 x 242023.04 | — | 101.49 | |
| GST-1.0Latent dimensions (N x M)=8 x 162023.04 | — | 98.29 | |
| GST-1.0Latent dimensions (N x M)=16 x 122023.04 | — | 98.12 | |
| GST-1.0Latent dimensions (N x M)=64 x 82023.04 | — | 102.53 | |
| GST-1.0Latent dimensions (N x M)=10 x 302023.04 | — | 98.64 | |
| GST-1.0Latent dimensions (N x M)=AVG2023.04 | — | 104.25 | |
| ReinMaxbatch size=200, training steps=5 x 10^5, latent dimensions=32, categorical dimensions=642023.04 | — | 93.44 | |
| ReinMaxLatent dimensions (N x M)=8 x 42023.04 | — | 124.66 | |
| ReinMaxLatent dimensions (N x M)=4 x 242023.04 | — | 99.77 | |
| ReinMaxLatent dimensions (N x M)=8 x 162023.04 | — | 97.7 | |
| ReinMaxLatent dimensions (N x M)=16 x 122023.04 | — | 98.06 | |
| ReinMaxLatent dimensions (N x M)=64 x 82023.04 | — | 100.71 | |
| ReinMaxLatent dimensions (N x M)=10 x 302023.04 | — | 98.37 | |
| ReinMaxLatent dimensions (N x M)=AVG2023.04 | — | 103.21 | |
| RLOObatch size=200, training steps=5 x 10^5, latent dimensions=32, categorical dimensions=642023.04 | — | 104.03 | |
| STbatch size=200, training steps=5 x 10^5, latent dimensions=32, categorical dimensions=642023.04 | — | 116 | |
| STLatent dimensions (N x M)=8 x 42023.04 | — | 135.53 | |
| STLatent dimensions (N x M)=4 x 242023.04 | — | 112.03 | |
| STLatent dimensions (N x M)=8 x 162023.04 | — | 112.94 | |
| STLatent dimensions (N x M)=16 x 122023.04 | — | 113.31 | |
| STLatent dimensions (N x M)=64 x 82023.04 | — | 113.9 | |
| STLatent dimensions (N x M)=10 x 302023.04 | — | 112.63 | |
| STLatent dimensions (N x M)=AVG2023.04 | — | 116.72 | |
| STGSbatch size=200, training steps=5 x 10^5, latent dimensions=32, categorical dimensions=642023.04 | — | 97.32 | |
| STGSLatent dimensions (N x M)=8 x 42023.04 | — | 126.85 | |
| STGSLatent dimensions (N x M)=4 x 242023.04 | — | 101.32 | |
| STGSLatent dimensions (N x M)=8 x 162023.04 | — | 99.32 | |
| STGSLatent dimensions (N x M)=16 x 122023.04 | — | 100.09 | |
| STGSLatent dimensions (N x M)=64 x 82023.04 | — | 104 | |
| STGSLatent dimensions (N x M)=10 x 302023.04 | — | 99.63 | |
| STGSLatent dimensions (N x M)=AVG2023.04 | — | 105.2 |