[ad_1]
arXiv:2405.00385v1 Announce Kind: new
Summary: The Bayes coding algorithm for context tree supply is a profitable instance of Bayesian tree estimation in textual content compression in data idea. This algorithm supplies an environment friendly parametric illustration of the posterior tree distribution and actual updating of its parameters. We apply this algorithm to a clustering job in machine studying. Extra particularly, we apply it to Bayesian estimation of the tree-structured stick-breaking course of (TS-SBP) combination fashions. For TS-SBP combination fashions, solely Markov chain Monte Carlo strategies have been proposed up to now, however any variational Bayesian strategies haven’t been proposed but. On this paper, we suggest a variational Bayesian technique that has a subroutine much like the Bayes coding algorithm for context tree sources. We affirm its habits by a numerical experiment on a toy instance.
[ad_2]
Supply hyperlink
Home ML/Data science blogs Variational Bayesian Strategies for a Tree-Structured Stick-Breaking Course of Combination of Gaussians