Density Estimation on Fashion (test)
2.803NLL (bits/dim)STEER
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| STEERParam=400K, Iter=60K2022.03 | 2.803 | 147.197 | 12.405 | 1,308.82 | |
| FFJORDParam=400K, Iter=60K2022.03 | 2.806 | 87.845 | 7.01 | 811.4 | |
| TO-FLOWParam=400K, Iter=60K2022.03 | 2.807 | 63.482 | 5.415 | 513.79 | |
| Real NVPTraining Approach=Maximum Likelihood2020.07 | 2.85 | — | — | — | |
| GlowTraining Approach=Maximum Likelihood2020.07 | 2.95 | — | — | — | |
| GISTraining Approach=Iterative2020.07 | 3.22 | — | — | — | |
| Sparse PC2022.11 | 3.27 | — | — | — | |
| BitSwap2022.11 | 3.28 | — | — | — | |
| HCLT2022.11 | 3.34 | — | — | — | |
| GFTraining Approach=Maximum Likelihood2020.07 | 3.35 | — | — | — | |
| IDF2022.11 | 3.47 | — | — | — | |
| BB-ANS2022.11 | 3.66 | — | — | — | |
| McBits2022.11 | 3.72 | — | — | — | |
| RatSPN2022.11 | 4.29 | — | — | — | |
| RBIGTraining Approach=Iterative2020.07 | 4.46 | — | — | — | |
| ByPE-ConvHVAEArchitecture=CNN with two stochastic layers, IWAE samples=5000, Pseudocoreset size=5002021.07 | 20.58 | — | — | — | |
| ConvHVAEArchitecture=CNN with two stochastic layers, Prior=VampPrior, IWAE samples=5000, Pseudo-inputs size=5002021.07 | 20.59 | — | — | — | |
| ConvHVAEArchitecture=CNN with two stochastic layers, Prior=Exemplar prior, IWAE samples=5000, Exemplar size=5002021.07 | 20.62 | — | — | — | |
| ConvHVAEArchitecture=CNN with two stochastic layers, Prior=Gaussian, IWAE samples=50002021.07 | 20.76 | — | — | — | |
| ByPE-HVAEArchitecture=MLP with two stochastic layers, IWAE samples=5000, Pseudocoreset size=5002021.07 | 20.82 | — | — | — | |
| HVAEArchitecture=MLP with two stochastic layers, Prior=VampPrior, IWAE samples=5000, Pseudo-inputs size=5002021.07 | 20.83 | — | — | — | |
| ByPE-VAEArchitecture=MLP with two hidden layers, IWAE samples=5000, Pseudocoreset size=5002021.07 | 20.85 | — | — | — | |
| VAEArchitecture=MLP with two hidden layers, Prior=VampPrior, IWAE samples=5000, Pseudo-inputs size=5002021.07 | 20.87 | — | — | — | |
| HVAEArchitecture=MLP with two stochastic layers, Prior=Exemplar prior, IWAE samples=5000, Exemplar size=5002021.07 | 20.95 | — | — | — | |
| VAEArchitecture=MLP with two hidden layers, Prior=Exemplar prior, IWAE samples=5000, Exemplar size=5002021.07 | 21 | — | — | — | |
| HVAEArchitecture=MLP with two stochastic layers, Prior=Gaussian, IWAE samples=50002021.07 | 21.04 | — | — | — | |
| VAEArchitecture=MLP with two hidden layers, Prior=Gaussian, IWAE samples=50002021.07 | 21.43 | — | — | — |