AI Paper Atlas

Research map / Probabilistic and scientific modeling / Variational and generative sampling

Improving multiple-try Metropolis with local balancing

Philippe Gagnon, Florian Maire, Giacomo Zanella

JMLR 2023 · Probabilistic and scientific modeling › Variational and generative sampling · approach: Stochastic gradient MCMC · 4 citations (OpenAlex)

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 changedWhereMeasuredEffect
use locally balanced weight sqrt(pi(y)/pi(x)) instead of pi(y)/pi(x) in MTM, increasing N (ablation)MTM · 50-d Gaussian, start in tailsconvergence time: LB MTM convergence speed increases with N, GB MTM deterioratesImproved
use Barker weight x/(1+x) instead of sqrt(x) in LB MTM (ablation)LB MTM · 50-d Gaussian and Laplace, convergence timeconvergence timeMixed
use LB weight instead of GB weight in stationarity when optimally tuned (ablation)MTM · 50-d Gaussian, stationarityESJD: ell = 3.20 optimal for all algorithmsNo clear effect
increase number of candidates N with fixed ell in GB MTM (ablation)GB MTM · 50-d Gaussian, stationarityESJDHurt
increase N from 1 to 20 with LB MTM (ablation)LB MTM · Immunotherapy DDE posteriorburn-in convergence speed: roughly by a factor of 7Improved
use LB MTM (N=20) instead of MH given parallel costMH · Immunotherapy DDE posteriorwall-clock burn-in: factor of roughly 2Improved
increase N in GB MTM (ablation)GB MTM · Immunotherapy DDE posteriorconvergenceHurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy