Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
134 tokens/sec
GPT-4o
9 tokens/sec
Gemini 2.5 Pro Pro
47 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

AutoStep: Locally adaptive involutive MCMC (2410.18929v3)

Published 24 Oct 2024 in stat.CO, cs.LG, and stat.ML

Abstract: Many common Markov chain Monte Carlo (MCMC) kernels can be formulated using a deterministic involutive proposal with a step size parameter. Selecting an appropriate step size is often a challenging task in practice; and for complex multiscale targets, there may not be one choice of step size that works well globally. In this work, we address this problem with a novel class of involutive MCMC methods -- AutoStep MCMC -- that selects an appropriate step size at each iteration adapted to the local geometry of the target distribution. We prove that under mild conditions AutoStep MCMC is $\pi$-invariant, irreducible, and aperiodic, and obtain bounds on expected energy jump distance and cost per iteration. Empirical results examine the robustness and efficacy of our proposed step size selection procedure, and show that AutoStep MCMC is competitive with state-of-the-art methods in terms of effective sample size per unit cost on a range of challenging target distributions.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (42)
  1. A general perspective on the Metropolis–Hastings kernel. arXiv:2012.14881.
  2. A tutorial on adaptive MCMC. Statistics and Computing, 18(4):343–373.
  3. Atchadé, Y. F. (2006). An adaptive version for the Metropolis adjusted Langevin algorithm with a truncated drift. Methodology and Computing in Applied Probability, 8(2):235–254.
  4. Comprehensive benchmarking of Markov chain Monte Carlo methods for dynamical systems. BMC Systems Biology.
  5. Hit-and-run algorithms for generating multivariate distributions. Mathematics of Operations Research, 18(2):255–266.
  6. autoMALA: Locally adaptive Metropolis-adjusted Langevin algorithm. In International Conference on Artificial Intelligence and Statistics, volume 238, pages 4600–4608.
  7. Incorporating local step-size adaptivity into the No-U-Turn Sampler using Gibbs self tuning. arXiv: 2408.08259.
  8. GIST: Gibbs self-tuning for locally adaptive Hamiltonian Monte Carlo. arXiv:2404.15253.
  9. Handling sparsity via the horseshoe. In International Conference on Artificial Intelligence and Statistics, volume 5, pages 73–80.
  10. Air Markov chain Monte Carlo. arXiv:1801.09309.
  11. Efficient and generalizable tuning strategies for stochastic gradient MCMC. Statistics and Computing, 33(3):66.
  12. Hybrid Monte Carlo. Physics Letters B, 195(2):216–222.
  13. Markov chain Monte Carlo: Can we trust the third significant figure? Statistical Science, 23(2).
  14. Fixed-width output analysis for Markov chain Monte Carlo. Journal of the American Statistical Association, 101(476):1537–1547.
  15. Turing: A language for flexible probabilistic inference. In International Conference on Artificial Intelligence and Statistics, pages 1682–1690.
  16. Bayesian Data Analysis. CRC Press, 3rd edition.
  17. Riemann manifold Langevin and Hamiltonian Monte Carlo methods. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 73(2):123–214.
  18. Delayed rejection in reversible jump Metropolis–Hastings. Biometrika, 88(4):1035–1053.
  19. An adaptive Metropolis algorithm. Bernoulli, 7(2):223–242.
  20. Hastings, W. K. (1970). Monte Carlo sampling methods using Markov chains and their applications. Biometrika, 57(1):97–109.
  21. The No-U-Turn Sampler: Adaptively setting path lengths in Hamiltonian Monte Carlo. The Journal of Machine Learning Research, 15(1):1593–1623.
  22. Kleppe, T. S. (2016). Adaptive step size selection for Hessian-based manifold Langevin samplers. Scandinavian Journal of Statistics, 43(3):788–805.
  23. Livingstone, S. (2021). Geometric ergodicity of the random walk Metropolis with position-dependent proposal covariance. Mathematics, 9(4).
  24. Markov kernels local aggregation for noise vanishing distribution sampling. SIAM Journal on Mathematics of Data Science, 4(4):1293–1319.
  25. An adaptive approach to Langevin MCMC. Statistics and Computing, 22.
  26. Equation of state calculations by fast computing machines. The Journal of Chemical Physics, 21(6):1087–1092.
  27. Delayed rejection Hamiltonian Monte Carlo for sampling multiscale distributions. Bayesian Analysis, pages 1–28.
  28. Neal, R. M. (1996). Bayesian Learning for Neural Networks, volume 118 of Lecture Notes in Statistics. Springer New York, New York, NY, 1 edition.
  29. Neal, R. M. (2003). Slice sampling. The Annals of Statistics, 31(3):705–767.
  30. Metropolis-Hastings view on variational inference and adversarial training. arXiv:1810.07151.
  31. Involutive MCMC: A unifying framework. In International Conference on Machine Learning.
  32. Variable length trajectory compressible hybrid Monte Carlo. arXiv:1604.00889.
  33. On the half-Cauchy prior for a global scale parameter. Bayesian Analysis, 7(4):887–902.
  34. Weak convergence and optimal scaling of random walk Metropolis algorithms. Annals of Applied Probability, 7(1):110–120.
  35. General state space Markov chains and MCMC algorithms. Probability Surveys, 1:20–71.
  36. Optimal scaling of discrete approximations to Langevin diffusions. Journal of the Royal Statistical Society. Series B (Statistical Methodology), 60(1):255–268.
  37. Brownian dynamics as smart Monte Carlo simulation. The Journal of Chemical Physics, 69(10):4628–4633.
  38. Connectionist Bench (Sonar, Mines vs. Rocks). UCI Machine Learning Repository.
  39. Pigeons.jl: Distributed sampling from intractable distributions. arXiv:2308.09769.
  40. Tierney, L. (1998). A note on Metropolis–Hastings kernels for general state spaces. The Annals of Statistics, 8(1):1–9.
  41. Some adaptive Monte Carlo methods for Bayesian inference. Statistics in Medicine, 18:2507–2515.
  42. Sampling from multiscale densities with delayed rejection generalized Hamiltonian Monte Carlo. arXiv:2406.02741.

Summary

  • The paper introduces AutoStep MCMC that autonomously adjusts step sizes based on local geometry for optimal sampling in involutive MCMC methods.
  • The method preserves π-invariance, irreducibility, and aperiodicity while enhancing effective sample size per computational cost.
  • Empirical results demonstrate that AutoStep MCMC outperforms state-of-the-art techniques in Bayesian inference on multiscale models.

Overview of AutoStep: Locally Adaptive Involutive MCMC

The paper introduces AutoStep MCMC, a novel class of Markov chain Monte Carlo (MCMC) methodologies that optimize step size selection on a local scale for involutive MCMC methods. This approach is designed to address the difficulties associated with choosing a suitable global step size for complex, multiscale target distributions, offering adaptability by tuning the step size at each iteration in accordance with the local geometry of the target distribution.

Core Contributions and Methodology

AutoStep MCMC operates within the domain of involutive MCMC, leveraging deterministic involutive proposals characterized by a step size parameter. The main contribution of AutoStep is its ability to autonomously adjust this step size during runtime, guided by the local behavior of the target distribution, denoted as π\pi.

Theoretically, the authors establish that AutoStep MCMC maintains π\pi-invariance, irreducibility, and aperiodicity under mild conditions. The method utilizes an augmented space approach incorporating both state and step size as part of the proposal distribution, enabling dynamic adaptation while preserving necessary MCMC properties.

Empirical Results

The empirical assessments conducted demonstrate the competitiveness of AutoStep MCMC against state-of-the-art alternatives, particularly noted in terms of effective sample size per computational cost across various challenging distributions. The flexibility of the proposed method is evident as it adapts effectively to different local structures of the target distribution.

Technical Observations

AutoStep MCMC integrates a round-based procedure for parameter tuning, optimizing the initial step size, jitter, and mass matrix over a series of iterations. The paper details a comprehensive step size selection strategy grounded in both operational efficiency and theoretical soundness, avoiding excessive computational burdens associated with exact calculations by implementing a heuristic range-based strategy for acceptance.

Implications and Future Directions

The AutoStep framework presents a robust mechanism for locally adaptive sampling, promising enhancements in exploring target distributions with intricate structures. Theoretically, this approach could inspire further investigations into MCMC sampling efficiency metrics, potentially exploring asymptotic properties and variance analysis.

In terms of practical application, AutoStep MCMC could influence a range of domains requiring Bayesian inference over multiscale models, offering enhanced sampling efficiency and reduced human intervention in tuning parameters.

Overall, this work enriches the toolbox for adaptive MCMC methods, underscoring the impact of local adaptivity and theoretical integration in developing efficient sampling algorithms. Future work could explore broader classes of targets, extend theoretical insights, and harness AutoStep MCMC's adaptability to complex, high-dimensional problems.

X Twitter Logo Streamline Icon: https://streamlinehq.com