Research map / Probabilistic and scientific modeling / Variational and generative sampling
Particle Gibbs Split-Merge Sampling for Bayesian Inference in Mixture Models
Alexandre Bouchard-Côté, Arnaud Doucet, Andrew Roth
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 |
|---|---|---|---|
| increase number of particles in PGSM (ablation) | PGSM (unmixed) · S1 synthetic 2D Normal | V-measure and predictive likelihood | No clear effect |
| use informed anchor proposal (cluster or threshold informed) instead of uniform (ablation) | PGSM · S1 synthetic 2D Normal | V-measure and predictive likelihood | No clear effect |
| use informed anchor proposal instead of uniform in higher dimensions (ablation) | PGSM · Ber-0.5 and DIM128 | V-measure and predictive likelihood | No clear effect |
| resample after every SMC iteration (relative ESS threshold 1) versus adaptive threshold (ablation) | PGSM · S1 synthetic 2D Normal | V-measure and predictive likelihood | Hurt |
| PGSM with three anchors instead of two (ablation) | PGSM finite mixture, max 5 clusters · C1 synthetic | predictive log likelihood | No clear effect |
| PGSM instead of Gibbs sampler | finite Dirichlet mixture, max 5 clusters · C1 synthetic | predictive log likelihood | Improved |
| pure PGSM kernel instead of Gibbs or SAMS | DP mixture sampler · S1 and S2 synthetic 2D Normal | predictive log likelihood and V-measure | Improved |
| pure PGSM kernel instead of Gibbs | DP mixture sampler · S3 and S4 synthetic 2D Normal | predictive log likelihood and V-measure | No clear effect |
| mix split-merge moves with Gibbs moves (ablation) | PGSM and SAMS · S1 and S2 synthetic 2D Normal | predictive log likelihood and V-measure | Improved |
| mix SAMS with Gibbs moves (ablation) | SAMS · 50-dim Bernoulli datasets | predictive likelihood and V-measure | Improved |
| unmixed PGSM instead of mixed PGSM with Gibbs (ablation) | PGSM · DIM128 Normal 128 dimensions | predictive likelihood and V-measure | Improved |
| PGSM or SAMS instead of Gibbs | DP mixture sampler · DIM064 and DIM128 Normal | predictive likelihood and V-measure | Improved |
Closest papers by meaning
- Parameter elimination in particle Gibbs sampling (NeurIPS 2019 · similarity 0.89)
- Particle-Gibbs Sampling for Bayesian Feature Allocation Models (JMLR 2021 · similarity 0.89)
- Efficient MCMC Sampling with Dimension-Free Convergence Rate using ADMM-type Splitting (JMLR 2022 · similarity 0.89)
- Diffusive Gibbs Sampling (ICML 2024 · similarity 0.85)
- To discretize continually: Mean shift interacting particle systems for Bayesian inference (NeurIPS 2026 · similarity 0.84)
- Split Gibbs Discrete Diffusion Posterior Sampling (NeurIPS 2025 · similarity 0.84)
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.
