Research map / Probabilistic and scientific modeling / Variational and generative sampling
Ensuring Rapid Mixing and Low Bias for Asynchronous Gibbs Sampling
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 |
|---|---|---|---|
| asynchronous (Hogwild!) execution of Gibbs sampling instead of sequential (ablation) | Gibbs sampling on a two-variable example model · two-variable example model | total variation distance to target: 9.8% TV distance, bias does not vanish with more samples | Hurt |
| measure bias with 1-sparse variation distance instead of total variation | Hogwild!-Gibbs bias evaluation · two-variable example model | 1-sparse vs total variation distance: 0.4% sparse vs 9.8% TV | No clear effect |
| increase expected delay tau* (asynchronicity) (ablation) | Hogwild!-Gibbs on synthetic Ising model · Ising n=1000, Delta=3 | estimated mixing time | Hurt |
| asynchronous multi-threaded Hogwild!-Gibbs execution | Gibbs sampling on TAC-KBP 11 GB dataset · TAC-KBP 11 GB | wall-clock speedup: up to 2.8x | Helped |
| Hogwild! vs multi-model Gibbs (independent chains) on large dataset | Gibbs sampling on TAC-KBP 11 GB dataset · TAC-KBP 11 GB | wall-clock run time | No clear effect |
| asynchronous execution on mixing-time adversarial example model (race conditions generate |1^T X| != 1) (ablation) | Gibbs sampling on constructed mixing-time example · constructed model, N=2001 | mixing time: exp(Omega(N)) vs O(N log N) sequential | Hurt |
| Dobrushin's condition (alpha<1) enforced: asynchronous mixing time bound | Hogwild!-Gibbs on models satisfying Dobrushin's condition · theory | mixing time bound ratio: factor 1+O(1/n) | No clear effect |
Closest papers by meaning
- Diffusive Gibbs Sampling (ICML 2024 · similarity 0.87)
- Composing Diffusion Priors with Explicit Physical Context via Generative Gibbs Sampling (NeurIPS 2026 · similarity 0.86)
- Multi-Marginal Couplings for Metropolis--Hastings (NeurIPS 2026 · similarity 0.86)
- Wasserstein Gradient Flows and Forward-Only Diffusion Are Not Enough for Multimodal Sampling (NeurIPS 2026 · similarity 0.86)
- Parallelizing MCMC Across the Sequence Length (NeurIPS 2025 · similarity 0.85)
- Parallel Affine Transformation Tuning of Markov Chain Monte Carlo (ICML 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.