AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Balancing Understanding and Generation in Discrete Diffusion Models

Yue Liu, Yuzhong Zhao, Zheyong Xie, Qixiang Ye, Jianbin Jiao, Yao Hu, Shaosheng Cao, Yunfan Liu

ICML 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
mix uniform noise into masked diffusion (XDLM vs UDLM) on likelihoodXDLM · OpenWebText validationperplexity: 25.937 -> 24.097Improved
XDLM vs MDLM on in-domain likelihoodXDLM · OpenWebText validationperplexity: 23.321 -> 24.097Hurt
XDLM vs UDLM on zero-shot average perplexityXDLM · Zero-shot, 7 datasetsperplexity: 59.574 -> 54.110Improved
XDLM vs MDLM on zero-shot average perplexityXDLM · Zero-shot, 7 datasetsperplexity: 53.650 -> 54.110No clear effect
XDLM vs MDLM, few-step image generation without CFGXDLM · ImageNet-1K, 4 steps, no CFGFID: 80.752 -> 54.085Improved
XDLM vs MDLM, 16 steps with CFGXDLM · ImageNet-1K, 16 steps, CFG 2.0FID: 6.725 -> 8.625Hurt
XDLM vs UDLM, few-step image generation with CFGXDLM · ImageNet-1K, 4 steps, CFG 2.0FID: 14.055 -> 13.550Improved
XDLM vs UDLM, 16 steps with CFGXDLM · ImageNet-1K, 16 steps, CFG 2.0FID: 8.980 -> 8.625Improved
continual pretraining of LLaDA with XDLM formulation (600 steps)LLaDA-8B · MBPPscore: 6.8 -> 15.0Improved
continual pretraining with standard MDLM objective instead of XDLM (ablation)LLaDA-8B · OpenCompass benchmarksscoreHurt
use XDLM-trained model instead of only XDLM sampling at inference (ablation)LLaDA-8B · GSM8k, MATH, BBH, MBPPscoreImproved
use mixing ratio k=0.1 between uniform and masked noise (ablation)XDLM · OWT generation vs zero-shot PPLPareto trade-offImproved

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