AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Understanding Diffusion Objectives as the ELBO with Simple Data Augmentation

Diederik P Kingma, Ruiqi Gao

NeurIPS 2023 · Image generation › Fast diffusion sampling · approach: Diffusion theory and design space · 11 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
Replace the Simple Diffusion weighting with shifted EDM-monotonic weighting and use the adaptive training scheduleSimple Diffusion U-ViT, L · ImageNet 128x128, no guidanceeval FID: 3.23 -> 2.88Improved
Use shifted sigmoid(-λ + 2) weighting and the adaptive training schedule instead of the baseline weighting and scheduleSimple Diffusion U-ViT, L · ImageNet 128x128, no guidanceInception Score: 171.9 ± 2.46 -> 183.1 ± 2.20Improved
Use EDM-monotonic weighting and the adaptive training scheduleSimple Diffusion U-ViT, L · ImageNet 128x128, guidanceeval FID: 3.57 -> 3.16Improved
Use EDM-monotonic weighting and the adaptive training scheduleSimple Diffusion U-ViT, L · ImageNet 256x256, no guidanceeval FID: 3.75 -> 3.36Improved
Use EDM-monotonic weighting and the adaptive training scheduleSimple Diffusion U-ViT, L · ImageNet 512x512, no guidanceeval FID: 4.53 -> 4.09Improved
Replace the non-monotonic sech(λ/2) weighting with sigmoid(-λ + 2)ε-prediction U-Net · ImageNet 64x64, DDPM samplerFID: 1.85 -> 1.68Improved
Use sigmoid(-λ + 5) instead of sech(λ/2)ε-prediction U-Net · ImageNet 64x64, DDPM samplerFID: 1.85 -> 1.94Hurt
Replace the EDM training schedule with the adaptive schedule (ablation)EDM parameterization · ImageNet 64x64, EDM samplerFID: 1.45 -> 1.43Improved
Replace EDM weighting with sigmoid(-λ + 2) weighting (ablation)EDM parameterization with adaptive schedule · ImageNet 64x64, EDM samplerFID: 1.43 -> 1.55Hurt
Replace the original v-prediction weighting exp(-λ/2) with EDM-monotonic weighting (ablation)v-prediction U-Net with adaptive schedule · ImageNet 64x64, EDM samplerFID: 1.62 -> 1.45Improved
Replace the baseline exp(-λ/2) weighting with shifted EDM-monotonic weighting (ablation)Simple Diffusion U-ViT, L · ImageNet 128x128, no guidanceInception Score: 171.9 ± 2.46 -> 171.1 ± 2.67No clear effect

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