Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Stacking for Non-mixing Bayesian Computations: The Curse and Blessing of Multimodal Posteriors
Yuling Yao, Aki Vehtari, Andrew Gelman
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 |
|---|---|---|---|
| stack 30 parallel chains instead of picking the best single chain | LDA topic model · Pride and Prejudice, test data | mean log predictive density: improvement in mean lpd (≈0.2) | Improved |
| stack chains instead of using BMA, pseudo-BMA or uniform averaging (ablation) | GP regression with Student-t likelihood · synthetic outliers, 8 chains | test log predictive density | Improved |
| use stacking weights instead of mode height or importance weighting (ablation) | GP hyperparameter mode combination · Neal dataset, n=100 | test log predictive density | Improved |
| stack complete pooling and centered chains instead of centered or non-centered alone | hierarchical model · synthetic, 300 test points | test log predictive density | Improved |
| stack 300 ADVI runs instead of selecting one run by ELBO or uniform mixing (ablation) | horseshoe logistic regression · synthetic, n=40, 400 predictors | test log predictive density | Improved |
| increase MCMC iterations from 2000 to 4000 per chain (ablation) | stacked LDA · L from 3 to 10 topics | test log predictive density: stacked average remains nearly unchanged | No clear effect |
| use pseudo-BMA or uniform weighting with more VI runs (ablation) | leukemia horseshoe VI · leukemia, 200 ADVI runs | leave-one-out lpd | Hurt |
Closest papers by meaning
- Scalable Nonparametric Sampling from Multimodal Posteriors with the Posterior Bootstrap (ICML 2019 · similarity 0.90)
- Stochastic Gradient Monomial Gamma Sampler (ICML 2017 · similarity 0.87)
- Pseudo-Extended Markov chain Monte Carlo (NeurIPS 2019 · similarity 0.87)
- Variational Inference with Mixtures of Isotropic Gaussians (NeurIPS 2025 · similarity 0.86)
- Scalable Bayesian Learning with posteriors (ICLR 2025 · similarity 0.85)
- Efficient Analytic Uncertainty Quantification for Multimodal Regression (NeurIPS 2026 · 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.
