Research map / Image generation / Fast diffusion sampling
Alternating Diffusion for Proximal Sampling with Zeroth Order Queries
Hirohane Takagi, Atsushi Nitanda
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 |
|---|---|---|---|
| enable interactions among N=100 particles instead of independent parallel chains (N=1 with 100 chains) (ablation) | Alternating diffusion proximal sampler · Gaussian Lasso mixture (d=5) | KL divergence convergence | Improved |
| use a 13.5 times larger step size h=1/10 with diffusion approximation instead of RGO with h=1/135 | proximal sampler · Gaussian Lasso mixture (d=5) | iterations to comparable KL: ~950 -> ~100 iterations | Improved |
| use the diffusion-based sampler instead of In-and-Out on a two-tori domain | proximal sampler for uniform distributions · Uniform on two disjoint tori | mode coverage | Improved |
Closest papers by meaning
- Beyond Scores: Proximal Diffusion Models (NeurIPS 2025 · similarity 0.88)
- Reverse Diffusion Sequential Monte Carlo Samplers (NeurIPS 2025 · similarity 0.88)
- Faster Sampling via Stochastic Gradient Proximal Sampler (ICML 2024 · similarity 0.88)
- Zeroth-Order Sampling Methods for Non-Log-Concave Distributions: Alleviating Metastability by Denoising Diffusion (NeurIPS 2024 · similarity 0.87)
- Wasserstein Gradient Flows and Forward-Only Diffusion Are Not Enough for Multimodal Sampling (NeurIPS 2026 · similarity 0.87)
- Diffusion Path Samplers via Sequential Monte Carlo (NeurIPS 2026 · similarity 0.87)
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.
