AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Continuously Augmented Discrete Diffusion model for Categorical Generative Modeling

Huangjie Zheng, Shansan Gong, Ruixiang ZHANG, Tianrong Chen, Jiatao Gu, Mingyuan Zhou, Navdeep Jaitly, Yizhe Zhang

ICLR 2026 · Image generation › Autoregressive and latent generation · approach: Latent diffusion with learned tokenizers · 27 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
augment masked diffusion with paired continuous latent (CADD)MDM · CIFAR-10FID: 4.66 -> 2.88Improved
augment masked diffusion with continuous latentMDM · ImageNet 32x32FID: 7.91 -> 3.74Improved
use CADD loss instead of MDM loss for code diffusion LMDiffuCoder · HumanEvalpass@1: 67.1 -> 72.0Improved
use CADD loss instead of MDM loss for code diffusion LMDiffuCoder · BigCodeBench-Hardpass@1: 12.8 -> 17.6Improved
use CADD loss instead of MDM lossDiffuCoder · MBPPpass@1: 74.2 -> 75.7Improved
fine-tune DiffuCoder checkpoint with CADD lossDiffuCoder · MBPPpass@1: 74.2 -> 73.9Hurt
fine-tune DiffuCoder checkpoint with CADD lossDiffuCoder · HumanEvalpass@1: 67.1 -> 73.8Improved
increase sampling steps from 1024 to 4096 (ablation)CADD · OpenWebText generationgenerative perplexity: 44.6 -> 35.3Improved
use CADD instead of MDLM on LM1B discrete perplexityMDLM · LM1Btest perplexity: 31.8 -> 31.4Improved
use CADD instead of RDLM on Text8RDLM · Text8bits per character: 1.32 -> 1.35Hurt
use CADD vs MDLM zero-shot perplexityMDLM · 7 zero-shot datasetsperplexity: CADD better on 4/7 tasksMixed
fine-tune MDLM checkpoint with CADD objective for 50B tokensMDLM · OpenWebText generationMAUVE and generative perplexityImproved

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