AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Maximum Likelihood Training of Score-Based Diffusion Models

Yang Song, Conor Durkan, Iain Murray, Stefano Ermon

NeurIPS 2021 · Image generation › Fast diffusion sampling · approach: Diffusion theory and learned samplers · 46 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Train with likelihood weighting instead of the original weighting (ablation)Score-based diffusion models · CIFAR-10NLL: 3.04 -> 2.94Improved
Train with likelihood weighting instead of the original weighting (ablation)Score-based diffusion models · ImageNet 32×32NLL: 3.84 -> 3.86Hurt
Add importance sampling to likelihood-weighted training (ablation)Score-based diffusion models · CIFAR-10loss variance after convergence: 98.48 -> 0.068Improved
Add importance sampling to likelihood-weighted training (ablation)Score-based diffusion models · ImageNetloss variance after convergence: 0.51 -> 0.043Improved
Combine likelihood weighting with importance sampling (ablation)Score-based diffusion models · CIFAR-10 and ImageNet 32×32NLLImproved
Use variational rather than uniform dequantization (ablation)Score-based diffusion models · CIFAR-10 and ImageNet 32×32NLL: 3.16 -> 3.04; 3.90 -> 3.84Improved
Use a deeper DDPM++ architecture (ablation)Score-based diffusion models · CIFAR-10NLL: 2.94 -> 2.93; 2.84 -> 2.81Mixed
Use a deeper DDPM++ architecture (ablation)Score-based diffusion models · ImageNet 32×32NLL: 3.80 -> 3.79; 3.79 -> 3.76Mixed
Use subVP rather than VP SDE with likelihood weighting and importance sampling (ablation)Score-based diffusion models · ImageNet 32×32NLL: 3.80 -> 3.79; 3.79 -> 3.76Improved
Use subVP rather than VP SDE with likelihood weighting and importance sampling (ablation)Score-based diffusion models · CIFAR-10NLL: 2.83 -> 2.84; 2.80 -> 2.81Mixed
Add variational dequantization to the ScoreFlow model (ablation)Score-based diffusion models · CIFAR-10 and ImageNet 32×32NLLImproved
Optimize likelihood using likelihood weighting and importance sampling (ablation)Score-based diffusion models · CIFAR-10FID: 3.98 -> 6.03Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy