[go: up one dir, main page]

Archive for ULA

scalable Monte Carlo for Bayesian learning [book review]

Posted in Books, Statistics, University life with tags , , , , , , , , , , , , , , , , , , , , on September 26, 2025 by xi'an

This book by Paul Fearnhead, Christopher Nemeth, Chris Oates, and Chris Sherlock is part of the IMS Monograph series. And published by Cambridge University Press. It covers most recent developments in MCMC methods, namely stochastic gradient MCMC (Chap. 3), non-reversible MCMC (Chap. 4), continuous-time MCMC (Chap. 5), and assessing and improving MCMC (Chap. 6). I find the book remarkable in its attention to rigour and clarity, without falling into overly technical derivations. It is perfectly suited for a graduate course to students with a solid mathematical background. In short, had I considered a new edition of our Monte Carlo Statistical Methods book to incorporate these advances, I could not done such a good job!

The first chapter provides a quick refresher of the background, from Monte Carlo principles, to Markov chains, SDEs, and the kernel “trick” (which requires a dozen pages of exposition). Nonetheless, it contains side remarks of true interest, including some suggestions I had not previously seen, as for instance an unusual introduction of the HMC algorithm as an underdamped Langevin diffusion. Chapter 2 prolongates this recap by covering reversible MCMC algorithms and the attached optimal scalings. This is done in a particularly friendly presentation that I intend to use in my own course. The HMC section is probably the best coverage I have seen on the topic, including most naturally the leapfrog steps.

Chapter 3 gets into stochastic gradient MCMC as an approximate MCMC, with nice arguments and formal convergence bounds. Again quite efficiently, if focussing almost solely on Gaussian settings (but including a neural network example). Similarly, Chapter 4 provides intuitive (if informal) arguments on the worth of non-reversible algorithms that are well-suited to a textbook of this level. This chapter introduces a PDMP sampler like the discrete bouncy particle sampler.

Chapter 5 is a (nicely) monstrous coverage of continuous time MCMC samplers that reaches very recent advances on PDMPs. The focus is on expressing them as limits, in order to derive mixing rates without extreme mathematical steps. (The chapter even includes a mention to the coordinate sampler that my PhD student Wu Changye derived in 2018!) Again a chapter I plan to use when teaching MCM methods, if possibly skipping some of the 66 pages.

Chapter 6 completes the monograph with a presentation of convergence assessment tools and diagnostics, exploiting the kernel trick, as well as convergence bounds that reflect very recent research in that domain. The conclusive section on optimal weights and optimal thinning will presumably be new to most readers. (Making me wonder if a link can be found with our importance Markov chain construct.)

[Disclaimer about potential self-plagiarism as usual: this post or an edited version will eventually appear in my Books Review section in CHANCE.]

Leave a comment »

proximal sampler

Posted in Books, pictures, R, Statistics, Travel, University life with tags , , , , , , , , , , , , on April 28, 2025 by xi'an

At the Columbia workshop last week, Andre Wibisono presented work related with a recent arXival on the exponentially fast convergence of both unadjusted Langevin and  proximal sampler algorithms under strong [definitely strong] log-concavity assumptions. The idea behind the proximal sampler is to target the demarginalised density

g(x,y) \propto \exp\{\log f(x) - ||x-y||^2/2\eta\}\quad\eta>0

by introducing an auxiliary Gaussian vector y, which preserves f(x) as the marginal distribution on the first component vector X. While the auxiliary Y is (obviously) conditionally Gaussian, the conditional of X is at least as challenging as simulating from f. Unless η is chosen small enough to regularize log g(x) into a strongly log-concave function, since

\log g(x,y) \le \log g(x^\star,y) -\beta||x-x^\star||^2

when x*=x*(y) is the maximiser of log g(x,y) (for a given value y) and β>0 is the appropriate log-concavity constant. This inequality means that an accept-reject can be implemented to simulate from the conditional of X given Y but it requires both the factor β and the derivation of x*(y), hence a pretty good understanding and a rather high regularity of the actual target f(x). Besides, the regularization term ||x-y||² means that y is approximately the previous value of the (sub)chain X, hence it creates a rappel force that slows down the exploration of the target.

Since the arXival does not contain numerical comparisons, I attempted one using the (2D) banana shaped distribution,

target=function(x,sig,B,mu)-x[1]^2/2/sig-(x[2]+B*x[1]^2-mu)^2/2

with μ=σ=B=10. Comparing with a vanilla random walk Metropolis with three potential scales, chosen randomly at each iteration. Since I did not want to check whether or not the target was log-concave (and derive the corresponding β), I used the Normal distribution centred at proposal x*(y) of a Metropolis step, again with several scales. The following is the representation of the samples (sienna for MCMC, navy blue for proximal with β=50, dark green for β=5), with a lesser rate of tail exploration for the proximal samplers. It is thus unclear to me the theoretical characterisations of the method translate into practical efficiency beyond the most regular cases.

1 Comment »