Research map / Image generation / Fast diffusion sampling
Proximal Diffusion Neural Sampler
Wei Guo, Jaemoo Choi, Yuchen Zhu, Molei Tao, Yongxin 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 resampling instead of importance weights (ablation) | PDNS · LJ-13 | energy W2: 5.96 -> 1.01 | Improved |
| use resampling instead of importance weights (ablation) | PDNS · DW-4 | energy W2: 0.32 -> 0.21 | Improved |
| use resampling instead of importance weights (ablation) | PDNS · DW-4 | W2: 0.50 -> 0.51 | No clear effect |
| use resampling instead of importance weights (ablation) | PDNS · Ising L=8 (beta=0.6) | adaptive training steps: 3080 -> 4339 | Hurt |
| use resampling instead of importance weights (ablation) | PDNS · Ising L=8 (beta=0.6) | energy W2: 1.052 -> 0.844 | Improved |
| add the proximal term to WDCE (ablation) | PDNS · multimodal targets | mode coverage | Improved |
| PDNS instead of LEAPS | PDNS · Potts L=16 q=4 (beta=1.3) | 2-point correlation error: 3.5e-1 -> 6.1e-4 | Improved |
| PDNS instead of MH | PDNS · Potts L=16 q=4 (beta=1.3) | 2-point correlation error: 6.4e-1 -> 6.1e-4 | Improved |
| PDNS instead of MH | PDNS · Ising L=24 (beta=0.4407) | 2-point correlation error: 5.6e-3 vs 1.9e-3 | Hurt |
| PDNS instead of NAAS | PDNS · MoS (d=50) | Sinkhorn: 394.55 -> 353.05 | Improved |
| PDNS instead of NAAS | PDNS · GMM40 (d=50) | Sinkhorn: 496.48 -> 327.83 | Improved |
| PDNS instead of ASBS | PDNS · LJ-55 (d=165) | energy W2: 27.69 -> 21.97 | Improved |
Closest papers by meaning
- MDNS: Masked Diffusion Neural Sampler via Stochastic Optimal Control (NeurIPS 2025 · similarity 0.90)
- Scalable Discrete Diffusion Samplers: Combinatorial Optimization and Statistical Physics (ICLR 2025 · similarity 0.88)
- Beyond Scores: Proximal Diffusion Models (NeurIPS 2025 · similarity 0.88)
- Learning to Discretize Denoising Diffusion ODEs (ICLR 2025 · similarity 0.88)
- On scalable and efficient training of diffusion samplers (NeurIPS 2025 · similarity 0.88)
- Progressive Tempering Sampler with Diffusion (ICML 2025 · 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.
