Research map / Image generation / Fast diffusion sampling
Generalization Bounds for Discrete Diffusion: Statistical Advantage of Masking
Zixuan Zhang, Hengyu Fu, Zhuoran Yang, Mengdi Wang, Tuo Zhao, Minshuo Chen
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use absorbing instead of uniform forward process at sparsity 10 (p_min=0.001) (ablation) | discrete diffusion · Synthetic Markov chain | normalized l1 distance: 35.38 -> 25.27 | Improved |
| use absorbing instead of uniform at sparsity 10 (p_min=0.004) (ablation) | discrete diffusion · Synthetic Markov chain | normalized l1 distance: 35.99 -> 22.66 | Improved |
| use absorbing instead of uniform at sparsity 20 (p_min=0.001) (ablation) | discrete diffusion · Synthetic Markov chain | normalized l1 distance: 36.49 -> 28.48 | Improved |
| use absorbing instead of uniform at sparsity 60 (p_min=0.001) (ablation) | discrete diffusion · Synthetic Markov chain | normalized l1 distance: 43.03 -> 43.01 | No clear effect |
| use absorbing instead of uniform at sparsity 80 (p_min=0.001) (ablation) | discrete diffusion · Synthetic Markov chain | normalized l1 distance: 45.17 -> 45.31 | No clear effect |
| use absorbing instead of uniform at sparsity 80 (p_min=0.004) (ablation) | discrete diffusion · Synthetic Markov chain | normalized l1 distance: 45.07 -> 45.61 | Hurt |
| increase p_min from 0.001 to 0.004 for uniform process at sparsity 10 (ablation) | uniform forward process · Synthetic Markov chain | normalized l1 distance: 35.38 -> 35.99 | No clear effect |
| increase p_min from 0.001 to 0.004 for absorbing process at sparsity 10 (ablation) | absorbing forward process · Synthetic Markov chain | normalized l1 distance: 25.27 -> 22.66 | Mixed |
| increase transition sparsity S from 10 to 80 (ablation) | absorbing process · Synthetic Markov chain | normalized l1 distance: 25.27 -> 45.31 | Hurt |
Closest papers by meaning
- Simplified and Generalized Masked Diffusion for Discrete Data (NeurIPS 2024 · similarity 0.93)
- Beyond Masked and Unmasked: Discrete Diffusion Models via Partial Masking (NeurIPS 2025 · similarity 0.91)
- Masked Diffusion Models are Secretly Time-Agnostic Masked Models and Exploit Inaccurate Categorical Sampling (ICLR 2025 · similarity 0.91)
- Accelerated Sampling from Masked Diffusion Models via Entropy Bounded Unmasking (NeurIPS 2025 · similarity 0.91)
- Simple and Effective Masked Diffusion Language Models (NeurIPS 2024 · similarity 0.91)
- Generalized Interpolating Discrete Diffusion (ICML 2025 · similarity 0.91)
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.
