Research map / Learning theory and algorithms / Sample complexity
Asymptotically Optimal Exact Minibatch Metropolis-Hastings
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 TunaMH instead of FlyMC | minibatch MH · Gaussian Mixture | symmetric KL versus wall-clock time | Helped |
| Use TunaMH instead of PoissonMH | minibatch MH · Gaussian Mixture | average batch size: 3969.67±327.26 -> 86.45±0.04 | Helped |
| Use TunaMH instead of FlyMC | minibatch MH · Logistic Regression | average batch size: 1960.19±150.96 -> 504.07±0.33 | Helped |
| Use TunaMH instead of TFMH | minibatch MH · Logistic Regression | wall-clock time to convergence | Helped |
| Increase χ (ablation) | TunaMH · MNIST Logistic Regression | convergence rate | Helped |
| Increase χ (ablation) | TunaMH · MNIST Logistic Regression | average batch size: 504.07 -> 810.35 -> 2047.91 | Hurt |
| Use first-order MAP control variates (SMH-1) with TFMH (ablation) | TFMH · Truncated Gaussian Mixture | sampling performance | Hurt |
| Add MAP-based control variates (ablation) | TFMH · Robust Linear Regression, d=100 | acceptance rate | Helped |
| Add MAP-based control variates (ablation) | TFMH · Robust Linear Regression, d=100 | average batch size | Hurt |
| Initialize the chain at the MAP solution (ablation) | TunaMH · Robust Linear Regression, d=100 | ESS per second | No clear effect |
| Use a local energy bound rather than PoissonMH's global bound | PoissonMH-style minibatch MH · Truncated Gaussian Mixture | average batch size: 3969.67±327.26 -> 86.45±0.04 | Helped |
| Use inexact minibatch MH rather than an exact method | minibatch MH · random walk example | stationary-distribution accuracy | Hurt |
Closest papers by meaning
- Scalable Metropolis-Hastings for Exact Bayesian Inference with Large Datasets (ICML 2019 · similarity 0.91)
- Rapidly Mixing Multiple-try Metropolis Algorithms for Model Selection Problems (NeurIPS 2022 · similarity 0.87)
- Multi-Marginal Couplings for Metropolis--Hastings (NeurIPS 2026 · similarity 0.87)
- Poisson-Minibatching for Gibbs Sampling with Convergence Rate Guarantees (NeurIPS 2019 · similarity 0.86)
- Near-Optimality of Contrastive Divergence Algorithms (NeurIPS 2024 · similarity 0.84)
- AutoStep: Locally adaptive involutive MCMC (ICML 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.