Research map / Graph learning / Graph self-supervised learning
DiGress: Discrete Denoising diffusion for graph generation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| replace Gaussian graph diffusion with discrete categorical diffusion (ablation) | DiGress versus ConGress · GuacaMol | FCD: 0.0 -> 68.0 | Helped |
| replace Gaussian graph diffusion with discrete categorical diffusion (ablation) | DiGress versus ConGress · QM9 | Training time (h): 7.2 -> 1.0 | Helped |
| use marginal transitions instead of uniform transitions (ablation) | DiGress · QM9 explicit hydrogens | Valid: $89.8 \spm{1.2}$ -> $92.3 \spm{2.5}$ | Helped |
| use marginal transitions instead of uniform transitions (ablation) | DiGress · QM9 explicit hydrogens | Mol stable: $70.5 \spm{2.1}$ -> $66.8 \spm{11.8}$ | Hurt |
| add spectral and structural features to marginal transitions (ablation) | DiGress · QM9 explicit hydrogens | Valid: $92.3 \spm{2.5}$ -> $\bm{95.4} \spm{1.1}$ | Helped |
| add spectral and structural features to marginal transitions (ablation) | DiGress · QM9 explicit hydrogens | Mol stable: $66.8 \spm{11.8}$ -> $\bm{79.8} \spm{5.6}$ | Helped |
| add marginal transitions and auxiliary features (ablation) | DiGress · QM9 explicit hydrogens | Unique: $97.8 \spm{0.2}$ -> $97.6 \spm{0.4}$ | No clear effect |
| use DiGress rather than SPECTRE | graph generation · Planar graphs | V.U.N.: 25\% -> 75\% | Helped |
| use DiGress rather than ConGress | graph generation · SBM | V.U.N.: 0\% -> 74\% | Helped |
| use DiGress rather than GraphINVENT | one-shot molecule generation · MOSES | FCD: 1.22 -> 1.19 | Helped |
| use DiGress rather than ConGress | one-shot molecule generation · GuacaMol | Valid: 0.1 -> 85.2 | Helped |
Closest papers by meaning
- Simple and Critical Iterative Denoising: A Recasting of Discrete Diffusion in Graph Generation (ICML 2025 · similarity 0.90)
- Graph Diffusion that can Insert and Delete (NeurIPS 2025 · similarity 0.90)
- Graph Diffusion Transformers for Multi-Conditional Molecular Generation (NeurIPS 2024 · similarity 0.89)
- Graph Generation with Diffusion Mixture (ICML 2024 · similarity 0.89)
- Is Noise Conditioning Necessary? A Unified Theory of Unconditional Graph Diffusion Models (NeurIPS 2025 · similarity 0.88)
- When Edge Independence Fails: Joint Graph Diffusion with Latent Sociability Priors (NeurIPS 2026 · 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.