Likelihood Estimation research benchmarks

Benchmarks

Dataset NameSOTA methodMetricTrendResultsLast Updated

CIFAR-10 (test)

i-DODE

2.42NLL (BPD)

Feb 26, 2026

MovieLens 10k (test)

NPwSA

-0.349Context Likelihood

Feb 26, 2026

ImageNet32 downsampled (test)

i-DODE

3.43BPD

Feb 26, 2026

CIFAR-10 original (test)

VDM

2.65BPD

Feb 26, 2026

CIFAR-10 uniformly dequantized (test)

sub-VP-ODE

3.02Bits/Dim

Feb 26, 2026

MNIST

NCP-VAE

78.1NLL

Feb 26, 2026

S&P 500 (d=98)

R-vine

-46.6Neg Log-Likelihood (-ℓ̄)

May 25, 2026

Arxiv Scientific Papers (val)

MDLM + Gumbel Distillation

16.85Perplexity

Mar 24, 2026

Pubmed Scientific Papers (val)

Block Diffusion + Gumbel Distillation

19.78Perplexity

Mar 24, 2026

Lambada (val)

MDLM + Gumbel Distillation

15.56Perplexity

Mar 24, 2026

PTB (val)

MDLM + Gumbel Distillation

35.12Perplexity

Mar 24, 2026

SVCEV model parameters {mu_i} (test)

Normalizing Flow (NF)

8.4Average Relative Error (Log-Likelihood)

Jun 4, 2026

SVCEV model at mu*

Normalizing Flow (NF)

0.04Relative Error of Log-Likelihood

Jun 4, 2026