Speaker
Description
In the era of cosmological analyses confronted with complex
distributions and a multitude of nuisance parameters, efficient
sampling methods are essential for accurate inference. This
presentation introduces the Approximate Posterior Ensemble Sampler
(APES), a novel algorithm designed to generate samples from challenging
target distributions that are traditionally difficult to handle using
Markov Chain Monte Carlo (MCMC) techniques. APES leverages kernel
density estimation and radial basis interpolation to construct an
adaptive proposal, enabling fast convergence of the chains and reduced
autocorrelation times. By comparing APES with the affine invariance
ensemble sampler using the stretch move, we demonstrate the superior
performance of APES in various contexts. Notably, on the Rosenbrock
function, APES achieves an impressive autocorrelation time 140 times
smaller than its counterpart. This presentation highlights the
practicality of APES as a scalable solution for sampling complex
distributions, offering a significant advantage for upcoming
cosmological surveys dealing with new systematics and an abundance of
nuisance parameters.
