AI & ML
Decomposition Bounds for MCMC with Mixture Targets: Multimodality and Heavy-Tailedness
arXiv stat.TH0 views
arXiv:2611.20513v2 Announce Type: replace Abstract: We establish quantitative mixing-time bounds for Markov chain Monte Carlo algorithms targeting multimodal and heavy-tailed mixtures. The dynamics of a Markov chain are decomposed into local and global levels, and possibly weak Poincar\'{e} inequalities at both levels are combined to form convergence bounds. We apply this framework to location and scale mixtures of strongly log-concave distributions, obtaining explicit convergence bounds for random-walk Metropolis, proximal, and random-scan Gibbs samplers. For location mixtures connected along a finite path, the bounds depend polynomially on the number of components and exponentially on neighboring separation, remaining informative as the overall diameter grows. For scale mixtures, we relate the global dynamics of the chain to the isoperimetry of the log-scale mixing distribution, and illustrate the effect of heavy-tailed targets by examining inverse-gamma mixing distributions.
Source:https://arxiv.org/abs/2611.20513PreviousSparse Fault Detection in Linear Systems: Equivalence of GLR Tests and $\ell_1$-regularized Least SquaresNextHigh-Dimensional Canonical Correlation Analysis
Related posts
How OpenAI’s Performance Factory keeps ChatGPT fast
0OpenAI Blog
Meta’s AI agent Muse is mixing up conversations and exposing private data, report says
0TechCrunch
Google gives Gemini the ability to use your Mac for you
0TechCrunch
Linkup /research is async. Treating it as synchronous is the bug, in code and in the freeze case
0dev.to
Fuck Your AI Safeguards
0Hacker News
Chrome AI features wasting much bandwidth
0Hacker News