AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Masked Auto-Regressive Variational Acceleration: Fast Inference Makes Practical Reinforcement Learning

Yuxuan Gu, Weimin Bai, Yifei Wang, Weijian Luo, He Sun

CVPR 2026 · Image generation › Fast diffusion sampling · approach: Masked diffusion language model decoding · 0 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
distill the diffusion chain into one step (GSIM), Base modelMAR-B · ImageNet 256x256inference time per image (s): 20.10 -> 0.61 (about 32.95x)Improved
distill the diffusion chain into one step, Base modelMAR-B · ImageNet 256x256FID: 2.60 -> 3.06Hurt
distill into one step, Huge modelMAR-H · ImageNet 256x256FID: 2.06 -> 2.00Improved
set distillation CFG scale to 1.2 instead of 1 (no CFG) (ablation)MARVAL-B · ImageNet 256x256FID: 3.75 -> 3.06Improved
raise distillation CFG scale from 1.2 to 4 (ablation)MARVAL-B · ImageNet 256x256FID: 3.06 -> 13.84Hurt
raise distillation CFG scale from 1.2 to 4 (ablation)MARVAL-B · ImageNet 256x256Inception Score: 220.2 -> 316.8Improved
add RL fine-tuning with PickScore reward (ablation)MARVAL (Base) · ImageNet class namesImageReward: -0.142 -> -0.110Improved
add RL fine-tuning with PickScore reward (ablation)MARVAL (Huge) · ImageNet class namesImageReward: -0.107 -> -0.064Improved
add RL fine-tuning with PickScore reward (ablation)MARVAL (Base) · ImageNet class namesCLIP score: 29.40 -> 29.47No clear effect
compress 20-step DDIM to 1 step and add RLDC-AR-512 · JourneyDB T2IImageReward: 0.5876 -> 0.6903Improved
couple RL directly with one-step GSIM samplesMARVAL · design argumentstabilityHurt

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