Research map / Optimization / Optimizers
Consistency and Fluctuations For Stochastic Gradient Langevin Dynamics
Yee Whye Teh, Alexandre H. Thiery, Sebastian J. Vollmer
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 exponent alpha=1/3 instead of smaller alpha (<1/3) (ablation) | SGLD · Linear Gaussian model | MSE decay rate: best rate of decay is 2/3 achieved at alpha=1/3 | Improved |
| use step-size exponent alpha>1/3 instead of alpha=1/3 (ablation) | SGLD · Linear Gaussian model | MSE decay rate: MSE scales as T_m^-1 for alpha>1/3 | Hurt |
| use SGLD instead of MALA when only a few passes through the data are allowed | Logistic regression (d=3, N=1000) · posterior variance estimate of first component | MSE | Improved |
| use SGLD instead of MALA asymptotically | Logistic regression · posterior variance estimate | convergence rate: n^{-1/3} vs n^{-1/2} | Hurt |
Closest papers by meaning
- Exploration of the (Non-)Asymptotic Bias and Variance of Stochastic Gradient Langevin Dynamics (JMLR 2016 · similarity 0.97)
- Stochastic Gradient Richardson-Romberg Markov Chain Monte Carlo (NeurIPS 2016 · similarity 0.94)
- The promises and pitfalls of Stochastic Gradient Langevin Dynamics (NeurIPS 2018 · similarity 0.94)
- Decentralized Stochastic Gradient Langevin Dynamics and Hamiltonian Monte Carlo (JMLR 2021 · similarity 0.92)
- Variance Reduction in Stochastic Gradient Langevin Dynamics (NeurIPS 2016 · similarity 0.92)
- Analysis of Langevin Monte Carlo via Convex Optimization (JMLR 2019 · similarity 0.92)
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.
