Compositional Models: Iterative Structure Learning from Data

0531044 - ÚTIA 2022 RIV SG eng C - Konferenční příspěvek (zahraniční konf.)
Kratochvíl, Václav - Bína, Vladislav - Jiroušek, Radim - Lee, T. R.
Compositional Models: Iterative Structure Learning from Data.
Sensor Networks and Signal Processing. vol. 176. Singapore: Springer, 2021 - (Peng, S.; Favorskaya, M.; Chao, H.), s. 379-395. 2190-3018. ISBN 978-981-15-4916-8.
[Sensor Networks and Signal Processing (SNSP 2019) /2./. Hualien (TW), 19.11.2019-22.11.2019]
Grant ostatní: GA ČR(CZ) GA19-06569S; Akademie věd - GA AV ČR(CZ) MOST-04-18
Program: GA
Institucionální podpora: RVO:67985556
Klíčová slova: Compositional models * Structure learning * Decomposability * Likelihood-ratio * Test statistics
Obor OECD: Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)
http://library.utia.cas.cz/separaty/2020/MTR/kratochvil-0531044.pdf

Multidimensional probability distributions that are too large to be stored in computer memory can be represented by a compositional model - a sequence of low-dimensional probability distributions that when composed together try to faithfully estimate the original multidimensional distribution. The decomposition to the compositional model is not satisfactorily resolved. We offer an approach based on search traversal through the decomposable model class using likelihood-test statistics. The paper is a work sketch of the current research.
Trvalý link: http://hdl.handle.net/11104/0310094