Research map / Image generation / Fast diffusion sampling
Unified Discrete Diffusion for Categorical Data
Lingxiao Zhao, Xueying Ding, Lijun Yu, Leman Akoglu
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 |
|---|---|---|---|
| combine VLB and CE losses instead of CE only (ablation) | USD3 (discrete-time) · Lakh Pianoroll Piano | 3-gram Hellinger distance: 0.574 -> 0.552 | Improved |
| combine VLB and CE losses instead of VLB only (ablation) | USD3 (discrete-time) · Lakh Pianoroll Piano-P | 3-gram train-to-test ratio: 2.863 -> 3.083 | Improved |
| combine VLB and CE losses instead of CE only (ablation) | USD3 (continuous-time) · Lakh Pianoroll Piano-P | 3-gram train-to-test ratio: 2.637 -> 3.083 | Improved |
| use continuous-time instead of discrete-time training | USD3 · Lakh Pianoroll Piano | 3-gram Hellinger distance: 0.552 -> 0.543 | Improved |
| use CE only loss instead of VLB (ablation) | USD3 (continuous-time) · Lakh Pianoroll Piano | 3-gram Prop. of Out.: 0.270 -> 0.346 | Hurt |
| use USD3 loss instead of D3PM loss | discrete-time diffusion · VQCIFAR10 | FID: 18.08 -> 13.25 | Improved |
| add CE-only USD3 loss instead of combined VLB+CE (ablation) | USD3 (continuous-time) · VQCIFAR10 | FID: 13.09 -> 11.97 | Improved |
| use continuous-time instead of discrete-time training | USD3* · VQCIFAR10 | FID: 12.07 -> 14.03 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SDDM — SDDM: Score-Decomposed Diffusion Models on Manifolds for Unpaired Image-to-Image Translation (ICML 2023) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Discrete Diffusion Models: Novel Analysis and New Sampler Guarantees (NeurIPS 2025 · similarity 0.90)
- Continuous Diffusion Model for Language Modeling (NeurIPS 2025 · similarity 0.90)
- Convergence of Score-Based Discrete Diffusion Models: A Discrete-Time Analysis (ICLR 2025 · similarity 0.89)
- Simple Guidance Mechanisms for Discrete Diffusion Models (ICLR 2025 · similarity 0.89)
- Scalable Discrete Diffusion Samplers: Combinatorial Optimization and Statistical Physics (ICLR 2025 · similarity 0.88)
- Fast Sampling via Discrete Non-Markov Diffusion Models with Predetermined Transition Time (NeurIPS 2024 · similarity 0.88)
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.
