Research map / Probabilistic and scientific modeling / Variational and generative sampling
Improving Gibbs Sampler Scan Quality with DoGS
Ioannis Mitliagkas, Lester Mackey
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 |
|---|---|---|---|
| Optimize a systematic scan with DoGS for a full-distribution Ising target | Gibbs sampling · 10×10 Ising model | Dobrushin variation guarantee: two orders of magnitude | Improved |
| Use a systematic scan instead of a uniform random scan | Gibbs sampling · 10×10 Ising model | Dobrushin variation guarantee | Improved |
| Customize the scan with DoGS to target X1 | Gibbs sampling · 1-million-variable Ising model | sampling steps: 2×10^6 -> 16 | Improved |
| Use DoGS instead of a systematic scan for center-region segmentation | MFAM Gibbs sampling · MSRC image segmentation | Hamming error | Improved |
| Use DoGS instead of a systematic scan for center-region segmentation | MFAM Gibbs sampling · MSRC image segmentation | scan length: 620 -> 110 | Improved |
| Use DoGS instead of a systematic scan for center-region segmentation | MFAM Gibbs sampling · MSRC image segmentation | time to resample variables of interest: 1.2s -> 0.37s | Improved |
| Initialize DoGS with a systematic scan and iterate the optimization on its own output (ablation) | DoGS · 10×10 Ising model | Dobrushin variation guarantee | Improved |
| Use a single DoGS optimization pass rather than iterating optimization (ablation) | DoGS · Ising scan optimization | Dobrushin variation guarantee | No clear effect |
| Loosen the influence bound used by DoGS computations (ablation) | DoGS · 40×40 toroidal Ising model, marginal inference | marginal inference accuracy: 50% | Hurt |
| Use DoGS rather than a uniform random scan for MCMC maximum likelihood estimation | MCMC maximum likelihood estimation · 3×3 and 4×4 Ising models | parameter estimation error | Improved |
| Use DoGS rather than a systematic scan for marginal inference | Gibbs sampling · 40×40 Ising model, top-left variable | marginal bias | Improved |
| Stop DoGS optimization at the user-specified accuracy ε (ablation) | DoGS · 300×300 Ising model, X1 expectation | long-term bias | Hurt |
Closest papers by meaning
- Minibatch Gibbs Sampling on Large Graphical Models (ICML 2018 · similarity 0.86)
- Particle-Gibbs Sampling for Bayesian Feature Allocation Models (JMLR 2021 · similarity 0.86)
- Scan Order in Gibbs Sampling: Models in Which it Matters and Bounds on How Much (NeurIPS 2016 · similarity 0.86)
- Diffusive Gibbs Sampling (ICML 2024 · similarity 0.84)
- Split Gibbs Discrete Diffusion Posterior Sampling (NeurIPS 2025 · similarity 0.82)
- Efficient Mixture Learning in Black-Box Variational Inference (ICML 2024 · similarity 0.81)
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.
