Research map / Image generation / Fast diffusion sampling
Learning Permutation Distributions via Reflected Diffusion on Ranks
Sizhuang He, Yangtian Zhang, Shiyang Zhang, David van Dijk
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 soft-rank Brownian-bridge forward process instead of riffle shuffle (ablation) | cGPL reverse, x_{t-1} · MNIST sorting N=32 | accuracy: 0.5833 -> 0.6425 | Improved |
| add biaffine pointer to cGPL (ablation) | Soft-Rank Diffusion · MNIST sorting N=32 | accuracy: 0.6425 -> 0.7152 | Improved |
| add biaffine pointer to cGPL (ablation) | Riffle Shuffle forward · MNIST sorting N=32 | accuracy: 0.5833 -> 0.6404 | Improved |
| use x_{t-1} instead of x_0 parametrization (ablation) | Riffle Shuffle + GPL · MNIST sorting N=32 | accuracy: 0.0000 -> 0.5896 | Improved |
| use x_{t-1} instead of x_0 parametrization (ablation) | Riffle Shuffle + cGPL · MNIST sorting N=32 | accuracy: 0.0000 -> 0.5833 | Improved |
| replace SymmetricDiffusers with Pointer-cGPL Soft-Rank | long sequences · MNIST sorting N=100 | accuracy: 0.0000 -> 0.1922 | Improved |
| replace SymmetricDiffusers with Soft-Rank cGPL | TSP-50 · TSP-50 | optimality gap: 1.2618 -> 0.0078 | Improved |
| add pointer to Soft-Rank cGPL (ablation) | Soft-Rank Diffusion · TSP-50 | tour length: 5.7336 -> 5.7157 | Improved |
| add pointer to Soft-Rank cGPL (ablation) | Soft-Rank Diffusion · CIFAR-10 jigsaw 8x8 | accuracy: 0.0924 -> 0.2700 | Improved |
| add pointer to Soft-Rank cGPL (ablation) | Soft-Rank Diffusion · CIFAR-10 jigsaw 4x4 | accuracy: 0.8752 -> 0.8681 | No clear effect |
| use Gaussian instead of uniform reference distribution (ablation) | Soft-Rank cGPL · CIFAR-10 jigsaw 4x4 | accuracy: 86.67 -> 86.96 | No clear effect |
| use Gaussian instead of uniform reference distribution (ablation) | Soft-Rank Pointer-cGPL · CIFAR-10 jigsaw 4x4 | accuracy: 87.35 -> 86.94 | No clear effect |
Closest papers by meaning
- SymmetricDiffusers: Learning Discrete Diffusion on Finite Symmetric Groups (ICLR 2025 · similarity 0.91)
- Scalable Discrete Diffusion Samplers: Combinatorial Optimization and Statistical Physics (ICLR 2025 · similarity 0.87)
- Pard: Permutation-Invariant Autoregressive Diffusion for Graph Generation (NeurIPS 2024 · similarity 0.87)
- O(d/T) Convergence Theory for Diffusion Probabilistic Models under Minimal Assumptions (ICLR 2025 · similarity 0.86)
- Beyond Scores: Proximal Diffusion Models (NeurIPS 2025 · similarity 0.86)
- Adapting to Unknown Low-Dimensional Structures in Score-Based Diffusion Models (NeurIPS 2024 · similarity 0.85)
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.
