List your company

Density Estimation on MNIST (test)

-3,408.16NLL (bits/dim)VAE
-3,548.0224-2,603.9512-1,659.88-715.8088Oct 2, 2018Jan 10, 2020Apr 20, 2021Jul 29, 2022Nov 7, 2023Feb 14, 2025May 26, 2026

Evaluation results

MethodLinks
VAE2019.10-3,408.16———
PresGAN2019.10-1,432.5———
GANentropy regularization (lambda)=02019.10-1,423.39———
Residual Flow2019.060.97———
MintNetType=Normalized, Single NFE=✓2026.050.98———
FFJORDArchitecture=multi-scale convolutional architectures2018.100.99———
FFJORD2018.110.99———
FFJORD2019.060.99———
FFJORDTraining Approach=Maximum Likelihood2020.070.99———
FFJORDType=Normalized, Single NFE=✗2026.050.99———
stNCE-s+DSMType=Estimate, Single NFE=✓, Time=0.272026.051———
FFJORDParam=400K, Iter=60K2022.031.01779.6416.409750.67
STEERParam=400K, Iter=60K2022.031.024138.21212.3681,265.48
TO-FLOWParam=400K, Iter=60K2022.031.02646.3633.353396.81
CTSM-vType=Estimate, Single NFE=✗2026.051.03———
GlowArchitecture=multi-scale convolutional architectures2018.101.05———
FFJORDArchitecture=single flow with a convolutional encoder-decoder architecture2018.101.05———
Glow2018.111.05———
Glow2019.061.05———
i-ResNet2019.061.05———
GlowTraining Approach=Maximum Likelihood2020.071.05———
GLOW2022.051.05———
GlowType=Normalized, Single NFE=✓2026.051.05———
Real NVPArchitecture=multi-scale convolutional architectures2018.101.06———
Real NVP2018.111.06———
i-ResNet2018.111.06———
Real NVP2019.061.06———
Real NVPTraining Approach=Maximum Likelihood2020.071.06———
I-RESNET2022.051.06———
i-ResNetType=Normalized, Single NFE=✗2026.051.06———
TREType=Estimate, Single NFE=✗2026.051.09———
Sparse PC2022.111.14———
HCLT2022.111.2———
BitSwap2022.111.27———
GFTraining Approach=Maximum Likelihood2020.071.29———
TSMType=Estimate, Single NFE=✗2026.051.3———
GISTraining Approach=Iterative2020.071.34———
FCONV2 + NSFlayer_type=flowified convolutional, kernels=non-overlapping, activation=rational quadratic splines2022.051.35———
BB-ANS2022.111.39———
FCONV2layer_type=flowified convolutional, kernels=non-overlapping2022.051.41———
stNCE-sType=Estimate, Single NFE=✓, Time=0.132026.051.61———
RatSPN2022.111.67———
RBIGTraining Approach=Iterative2020.071.71———
MAF2018.101.89———
MAF2018.111.89———
MAFTraining Approach=Maximum Likelihood2020.071.89———
IDF2022.111.9———
MAF2022.051.91———
McBits2022.111.98———
MADE2018.102.04———
MADE2018.112.04———
FCONV1 + NSFlayer_type=flowified convolutional, kernels=overlapping, activation=rational quadratic splines2022.052.7———
FP-DiffusionType=Normalized, Single NFE=✗2026.052.98———
FCONV1layer_type=flowified convolutional, kernels=overlapping2022.053.11———
FMLPlayer_type=flowified linear2022.054.19———
NICE2018.114.36———
stNCE-wType=Estimate, Single NFE=✓, Time=0.132026.054.56———
stNCE-mType=Estimate, Single NFE=✓, Time=0.17*2026.054.92———
tNCEType=Estimate, Single NFE=✓, Time=0.1272026.057.01———
UHAK=642021.0786.3———
IWK=642021.0786.6———
IWK=322021.0786.8———
IWK=162021.0787———
UHAK=322021.0787.3———
IWK=82021.0787.3———
UHAK=162021.0787.4———
UHAK=82021.0787.6———
IWK=12021.0788.3———
UHAK=12021.0788.4———