Research map / Optimization / Optimizers
Exploration of the (Non-)Asymptotic Bias and Variance of Stochastic Gradient Langevin Dynamics
Sebastian J. Vollmer, Konstantinos C. Zygalakis, Yee Whye Teh
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 |
|---|---|---|---|
| add a bias-correcting term to SGLD (mSGLD) removing first-order gradient-variance bias | SGLD · Gaussian toy model, second moment | asymptotic bias / MSE2 | Improved |
| use mSGLD instead of SGLD when the number of steps is large (high accuracy regime) | SGLD · Gaussian toy model, optimal cost | computational cost for MSE2 | Improved |
| use mSGLD with small data batches (n = N^0.1, N^0.5) | SGLD · Gaussian toy model, growing N | MSE2 | Hurt |
| vary subset size for mSGLD in a Bayesian logistic-style example: n=150, n=50, n=10 (ablation) | mSGLD · 2D example, posterior mean, h = 0.002 | MSE | Mixed |
| subsample the data (SGLD) instead of full gradients (Euler) as N grows | Euler method · Gaussian toy model, N to infinity | computational cost for vanishing MSE2 | Improved |
Closest papers by meaning
- Consistency and Fluctuations For Stochastic Gradient Langevin Dynamics (JMLR 2016 · similarity 0.97)
- The promises and pitfalls of Stochastic Gradient Langevin Dynamics (NeurIPS 2018 · similarity 0.95)
- Stochastic Gradient Richardson-Romberg Markov Chain Monte Carlo (NeurIPS 2016 · similarity 0.94)
- Variance Reduction in Stochastic Gradient Langevin Dynamics (NeurIPS 2016 · similarity 0.92)
- Decentralized Stochastic Gradient Langevin Dynamics and Hamiltonian Monte Carlo (JMLR 2021 · similarity 0.92)
- On Stationary-Point Hitting Time and Ergodicity of Stochastic Gradient Langevin Dynamics (JMLR 2020 · 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.
