mlsanityData to 7 Oct 2026

Efficient and Generalizable Tuning Strategies for Stochastic Gradient MCMC

arXiv 2105.13059 ↗First posted 27 May 2021stat.CO stat.ME stat.MLcited by 11

Not on the map. Why?

Abstract (arXiv)

Stochastic gradient Markov chain Monte Carlo (SGMCMC) is a popular class of algorithms for scalable Bayesian inference. However, these algorithms include hyperparameters such as step size or batch size that influence the accuracy of estimators based on the obtained posterior samples. As a result, these hyperparameters must be tuned by the practitioner and currently no principled and automated way to tune them exists. Standard MCMC tuning methods based on acceptance rates cannot be used for SGMCMC, thus requiring alternative tools and diagnostics. We propose a novel bandit-based algorithm that tunes the SGMCMC hyperparameters by minimizing the Stein discrepancy between the true posterior and its Monte Carlo approximation. We provide theoretical results supporting this approach and assess various Stein-based discrepancies. We support our results with experiments on both simulated and real datasets, and find that this method is practical for a wide range of applications.

Key facts. Cited by 11 papers in the map (data through 7 Oct 2026). Builds on 22 of its 44 references that are in the map. First posted on arXiv 27 May 2021.

Builds on

22 of 44 references.

References · 22

  1. 2017On Calibration of Modern Neural Networkscited by 4,860
  2. 2016Simple and Scalable Predictive Uncertainty Estimation using Deep Ensemblescited by 3,888
  3. 2019Introduction to Multi-Armed Banditscited by 475
  4. 2018Pyro: Deep Universal Probabilistic Programmingcited by 379
  5. 2014Stochastic Gradient Hamiltonian Monte Carlocited by 341
  6. 2021What Are Bayesian Neural Network Posteriors Really Like?cited by 307
  7. 2016A Kernelized Stein Discrepancy for Goodness-of-fit Tests and Model Evaluationcited by 236
  8. 2015Non-stochastic Best Arm Identification and Hyperparameter Optimizationcited by 219
  9. 2019Composable Effects for Flexible and Accelerated Probabilistic Programming in NumPyrocited by 187
  10. 2015A Complete Recipe for Stochastic Gradient MCMCcited by 146
  11. 2016A Kernel Test of Goodness of Fitcited by 143
  12. 2015Preconditioned Stochastic Gradient Langevin Dynamics for Deep Neural Networkscited by 121
10 more
  1. 2017Measuring Sample Quality with Kernelscited by 119
  2. 2015Measuring Sample Quality with Stein's Methodcited by 94
  3. 2018The promises and pitfalls of Stochastic Gradient Langevin Dynamicscited by 56
  4. 2019Stochastic gradient Markov chain Monte Carlocited by 56
  5. 2017A Linear-Time Kernel Goodness-of-Fit Testcited by 41
  6. 2020Stochastic Stein Discrepanciescited by 30
  7. 2017Control Variates for Stochastic Gradient MCMCcited by 25
  8. 2020Sliced Kernelized Stein Discrepancycited by 23
  9. 2015Bridging the Gap between Stochastic Gradient MCMC and Stochastic Optimizationcited by 17
  10. 2020Stochastic Gradient Langevin Dynamics Algorithms with Adaptive Driftscited by 12

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 1; 2023: 1; 2024: 5; 2025: 1; 2026: 3.

11 papers.

  1. 2023Tuning-Free Maximum Likelihood Training of Latent Variable Models via Coin Bettingcited by 11
  2. 2026Can Microcanonical Langevin Dynamics Leverage Mini-Batch Gradient Noise?cited by 6
  3. 2022Tuning Stochastic Gradient Algorithms for Statistical Inference via Large-Sample Asymptoticscited by 4
  4. 2024Reinforcement Learning for Adaptive MCMCcited by 4
  5. 2024Nystr"om Kernel Stein Discrepancycited by 4
  6. 2024AutoStep: Locally adaptive involutive MCMCcited by 4
  7. 2024Robust Approximate Sampling via Stochastic Gradient Barker Dynamicscited by 2
  8. 2025Harnessing the Power of Reinforcement Learning for Adaptive MCMCcited by 2
  9. 2024Dynamic Learning Rate for Deep Reinforcement Learning: A Bandit Approachcited by 0
  10. 2026Neural-Inspired Posterior Approximation (NIPA)cited by 0
  11. 2026Large-scale Uncertainty Quantification for Latent Variable Models Using Subsampling Markov Chain Monte Carlocited by 0