AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Censored Sampling of Diffusion Models Using 3 Minutes of Human Feedback

TaeHo Yoon, Kibeom Myoung, Keon Lee, Jaewoong Cho, Albert No, Ernest K. Ryu

NeurIPS 2023 · Image generation › Fast diffusion sampling · approach: Diffusion models as perception tools · 2 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
use reward ensemble plus universal guidance instead of single reward model (ablation)censoring pipeline · MNIST crossed 7proportion of malign images (%): 1.30 -> 0.42Improved
use ensemble instead of union reward model (ablation)censoring pipeline · MNIST crossed 7proportion of malign images (%): 0.98 -> 0.60Improved
add universal guidance (backward + recurrence) to ensemble (ablation)censoring pipeline · MNIST crossed 7proportion of malign images (%): 0.60 -> 0.42Improved
use ensemble instead of single reward model (ablation)censoring pipeline · LSUN church watermarkproportion of malign images (%): 3.02 -> 1.36Improved
add recurrence to ensemble (ablation)censoring pipeline · LSUN church watermarkproportion of malign images (%): 1.36 -> 0.76Improved
single models trained on 30 malign + 30 benign vs union models (ablation)reward model · LSUN church watermarkproportion of malign images (%): 3.84 -> 3.02Improved
use imitation learning (3 rounds) instead of training at once on 30 malign (ablation)reward model · ImageNet tench facesproportion of malign images (%): 6.8 -> 2.2Improved
use imitation learning (2 rounds) instead of training at once on 20 malign (ablation)reward model · ImageNet tench facesproportion of malign images (%): 10.7 -> 7.5Improved
add universal guidance to imitation learning round 3 (ablation)censoring pipeline · ImageNet tench facesproportion of malign images (%): 2.2 -> 1.0Improved
use guidance-based censoring instead of rejection sampling (threshold 0.8) (ablation)censoring pipeline · ImageNet tench facesproportion of malign images (%): 29.8 -> 2.2Improved
use ensemble + universal guidance instead of ensemble only (ablation)censoring pipeline · Stable Diffusion embedded textproportion of malign images (%): 3.86 -> 1.24Improved
use ensemble instead of single reward model (ablation)censoring pipeline · Stable Diffusion embedded textproportion of malign images (%): 7.10 -> 3.86Improved

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