Previous |  Up |  Next

Article

Title: On a result of K. P. Hart about non-existence of measurable solutions to the discrete expectation maximization problem (English)
Author: Pestov, Vladimir G.
Language: English
Journal: Commentationes Mathematicae Universitatis Carolinae
ISSN: 0010-2628 (print)
ISSN: 1213-7243 (online)
Volume: 64
Issue: 3
Year: 2023
Pages: 353-358
Summary lang: English
.
Category: math
.
Summary: It was shown that there is a statistical learning problem -- a version of the expectation maximization (EMX) problem -- whose consistency in a domain of cardinality continuum under the family of purely atomic probability measures and with finite hypotheses is equivalent to a version of the continuum hypothesis, and thus independent of ZFC. K. P. Hart had subsequently proved that no solution to the EMX problem can be Borel measurable with regard to an uncountable standard Borel structure on $X$, and so the independence result could just be an artefact of a model allowing non-measurable learning rules. In this note we reinforce the point somewhat by observing that such a solution cannot even be Lebesgue measurable. (English)
Keyword: expectation maximization problem
Keyword: EMX
Keyword: continuum hypothesis
Keyword: independence of ZFC
Keyword: measurability
MSC: 03E35
MSC: 68T05
DOI: 10.14712/1213-7243.2023.031
.
Date available: 2024-03-18T10:44:16Z
Last updated: 2024-03-18
Stable URL: http://hdl.handle.net/10338.dmlcz/152303
.
Reference: [1] Ben-David S., Hrubeš P., Moran S., Shpilka A., Yehudayoff A.: A learning problem that is independent of the set theory ZFC axioms.available at arXiv 1711.05195v1 [cs.LG] (2017), 17 pages.
Reference: [2] Ben-David S., Hrubeš P., Moran S., Shpilka A., Yehudayoff A.: Learnability can be undecidable.Nat. Mach. Intell. 1 (2019), no. 1, 44–48. MR 4233057, 10.1038/s42256-018-0002-3
Reference: [3] Hart K. P.: Machine learning and the continuum hypothesis.Nieuw Arch. Wiskd. (5) 20 (2019), no. 3, 214–217. MR 4448792
.

Fulltext not available (moving wall 24 months)

Partner of
EuDML logo