Single MCMC chain parallelisation on decision trees

Drousiotis, Efthyvoulos ORCID: 0000-0002-9746-456X and Spirakis, Paul ORCID: 0000-0001-5396-3749 (2023) Single MCMC chain parallelisation on decision trees. ANNALS OF MATHEMATICS AND ARTIFICIAL INTELLIGENCE. pp. 1-14.

Text
Annals_of_Mathematics_and_Artificial_Intelligence (4).pdf - Author Accepted Manuscript
Download (386kB) | Preview

Official URL: http://dx.doi.org/10.1007/s10472-023-09876-9

Abstract

<jats:title>Abstract</jats:title><jats:p>Decision trees (DT) are highly famous in machine learning and usually acquire state-of-the-art performance. Despite that, well-known variants like CART, ID3, random forest, and boosted trees miss a probabilistic version that encodes prior assumptions about tree structures and shares statistical strength between node parameters. Existing work on Bayesian DT depends on Markov Chain Monte Carlo (MCMC), which can be computationally slow, especially on high dimensional data and expensive proposals. In this study, we propose a method to parallelise a single MCMC DT chain on an average laptop or personal computer that enables us to reduce its run-time through multi-core processing while the results are statistically identical to conventional sequential implementation. We also calculate the theoretical and practical reduction in run time, which can be obtained utilising our method on multi-processor architectures. Experiments showed that we could achieve 18 times faster running time provided that the serial and the parallel implementation are statistically identical.</jats:p>

Item Type:	Article
Uncontrolled Keywords:	Parallel algorithms, Machine learning, MCMC decision tree
Divisions:	Faculty of Science and Engineering > School of Electrical Engineering, Electronics and Computer Science
Depositing User:	Symplectic Admin
Date Deposited:	02 Jun 2023 09:35
Last Modified:	15 Mar 2024 17:38
DOI:	10.1007/s10472-023-09876-9
Related URLs:	Author Publisher
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3170793