Research map / Learning theory and algorithms / Sample complexity
Minimax Mixing Time of the Metropolis-Adjusted Langevin Algorithm for Log-Concave Sampling
Keru Wu, Scott Schmidler, Yuansi Chen
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 |
|---|---|---|---|
| use step size h=d^-0.5 instead of larger h=d^-0.2 or d^-0.35 for MALA (ablation) | MALA · Worst-case target (warm start, L=m=1) | acceptance rate and mixing time proxy: h=d^-0.2 and d^-0.35 acceptance tends to zero at large d; d^-0.5 best | Improved |
| use step size smaller than d^-0.5 instead of d^-0.5 (ablation) | MALA · Worst-case target (warm start, L=m=1) | mixing time proxy: acceptance rate stays positive but mixing time larger than for h=d^-0.5 | Hurt |
| use step size d^-1 instead of d^-0.5 under an exponentially-warm Gaussian start (ablation) | MALA · Worst-case target (N(0, I/1000) start) | acceptance rate and mixing time proxy: d^-0.5 has close-to-zero acceptance; d^-1 best | Improved |
| use condition number dependency kappa^-1 instead of kappa^-0.5 or kappa^-0.75 in step size (ablation) | MALA · Worst-case target (d=32, warm start) | acceptance rate and mixing time proxy: kappa^-0.5 and kappa^-0.75 give vanishing acceptance and large mixing times | Improved |
Closest papers by meaning
- Log-concave sampling: Metropolis-Hastings algorithms are fast (JMLR 2019 · similarity 0.96)
- On the Computational Complexity of Metropolis-Adjusted Langevin Algorithms for Bayesian Posterior Sampling (JMLR 2024 · similarity 0.93)
- Lower Bounds on Metropolized Sampling Methods for Well-Conditioned Distributions (NeurIPS 2021 · similarity 0.92)
- Langevin Monte Carlo for strongly log-concave distributions: Randomized midpoint revisited (ICLR 2024 · similarity 0.90)
- 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 0.89)
- Provable Benefit of Annealed Langevin Monte Carlo for Non-log-concave Sampling (ICLR 2025 · similarity 0.89)
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.
