Research map / Learning theory and algorithms / Sample complexity
Log-concave sampling: Metropolis-Hastings algorithms are fast
Raaz Dwivedi, Yuansi Chen, Martin J. Wainwright, Bin Yu
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 |
|---|---|---|---|
| replace ULA with MALA (add accept-reject step) | ULA · Gaussian, strongly log-concave (kappa=4, delta=0.2) | error-dependence slope of mixing time: 2.23 -> 0.33 | Improved |
| replace ULA with MALA (add accept-reject step) | ULA · Gaussian, weakly log-concave (lambda_max=1000, lambda_min=1) | error-dependence slope of mixing time: 3.92 -> 1.56 | Improved |
| replace ULA with MALA | ULA · Gaussian, weakly log-concave (lambda_max=1000, lambda_min=1) | dimension-dependence slope of mixing time: 2.78 -> 1.61 | Improved |
| replace MRW with MALA | MRW · Gaussian, weakly log-concave (lambda_max=1000, lambda_min=1) | dimension-dependence slope of mixing time: 2.73 -> 1.61 | Improved |
| replace ULA with MALA or MRW | ULA · Gaussian, strongly log-concave (kappa=4, delta=0.2) | dimension-dependence slope of mixing time: 1.01 -> 0.84 | No clear effect |
| use MALA instead of ULA | ULA · 2D Gaussian mixture | discretized TV error vs iterations | Improved |
| shrink ULA step size via smaller delta (0.2 -> 0.1) | ULA · 2D Gaussian mixture | convergence rate | Hurt |
| enlarge ULA step size via larger delta (0.2 -> 1.0) | ULA · 2D Gaussian mixture | asymptotic TV error | Hurt |
| use MALA instead of ULA and MRW | ULA/MRW · 2D Gaussian mixture | trace-plot stabilization and autocorrelation | Improved |
| add preconditioning with Sigma_X^{-1/2} | ULA and MRW · Bayesian logistic regression (d=2, n=50) | mean error of posterior sample mean | Improved |
| use MALA instead of ULA | ULA · Bayesian logistic regression (d=2, n=50) | mean error vs iterations | Improved |
| increase MALA step size from d^-0.5 to d^-0.2 (ablation) | MALA · Gaussian and logistic regression posterior | average acceptance probability | Hurt |
Closest papers by meaning
- Langevin Monte Carlo for strongly log-concave distributions: Randomized midpoint revisited (ICLR 2024 · similarity 0.92)
- Chain of Log-Concave Markov Chains (ICLR 2024 · similarity 0.92)
- Non-asymptotic Error Bounds in $\mathcal{W}_2$-Distance with Sqrt(d) Dimension Dependence and First Order Convergence for Langevin Monte Carlo beyond Log-Concavity (ICML 2025 · similarity 0.91)
- Provable Benefit of Annealed Langevin Monte Carlo for Non-log-concave Sampling (ICLR 2025 · similarity 0.91)
- Faster Sampling from Log-Concave Densities over Polytopes via Efficient Linear Solvers (ICLR 2024 · similarity 0.90)
- Faster Diffusion Sampling with Randomized Midpoints: Sequential and Parallel (ICLR 2025 · similarity 0.90)
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.
