Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Scalable Metropolis-Hastings for Exact Bayesian Inference with Large Datasets
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 |
|---|---|---|---|
| Use SMH-2 with a second-order Taylor control variate | Metropolis-Hastings for Bayesian logistic regression · Bayesian logistic regression, increasing n | average likelihood evaluations per step: O(1/√n) | Helped |
| Use SMH-1 with a first-order Taylor control variate | Metropolis-Hastings for Bayesian logistic regression · Bayesian logistic regression, increasing n | average likelihood evaluations per step: O(1) | Helped |
| Use SMH-2 rather than competing methods | Bayesian logistic regression · Large n | ESS for first regression coefficient per execution time | Helped |
| Use a second-order rather than first-order Taylor approximation (ablation) | SMH · Bayesian logistic regression, increasing n | average likelihood evaluations per step: O(1) -> O(1/√n) | Helped |
| Use direct likelihood factorization instead of Taylor control variates (ablation) | Factorized Metropolis-Hastings · Bayesian logistic regression | overall performance: Θ(√n) | Hurt |
| Use a smaller proposal step size of O(1/n) rather than matching posterior concentration (ablation) | Naive factorized Metropolis-Hastings · Bayesian logistic regression | overall performance | Hurt |
| Use an independent pCN proposal (ρ=0) as the target concentrates (ablation) | SMH-2 · Bayesian logistic regression | ESS for first regression coefficient per execution time: ρ=0 | Helped |
| Use SMH rather than the Gaussian approximation alone | Gaussian approximation [0mπ̂₂ · Bayesian logistic regression, reasonably large n | accuracy | Helped |
| Use SMH rather than control-variate subsampling at very large n | Bayesian logistic regression · Very large n | practical usefulness | No clear effect |
Closest papers by meaning
- Asymptotically Optimal Exact Minibatch Metropolis-Hastings (NeurIPS 2020 · similarity 0.91)
- Large-Scale Stochastic Sampling from the Probability Simplex (NeurIPS 2018 · similarity 0.90)
- Multi-Marginal Couplings for Metropolis--Hastings (NeurIPS 2026 · similarity 0.90)
- Parallelizing MCMC Across the Sequence Length (NeurIPS 2025 · similarity 0.88)
- AutoStep: Locally adaptive involutive MCMC (ICML 2025 · similarity 0.88)
- Metropolis Adjusted Microcanonical Hamiltonian Monte Carlo (NeurIPS 2025 · similarity 0.88)
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.