Version 1
: Received: 14 February 2022 / Approved: 21 February 2022 / Online: 21 February 2022 (14:13:26 CET)
Version 2
: Received: 7 June 2023 / Approved: 7 June 2023 / Online: 7 June 2023 (13:20:18 CEST)
How to cite:
Gogoshin, G.; Rodin, A. Resolution Limit in Statistical Independence and Bayesian Network Scoring Functions. Preprints2022, 2022020254. https://doi.org/10.20944/preprints202202.0254.v1
Gogoshin, G.; Rodin, A. Resolution Limit in Statistical Independence and Bayesian Network Scoring Functions. Preprints 2022, 2022020254. https://doi.org/10.20944/preprints202202.0254.v1
Gogoshin, G.; Rodin, A. Resolution Limit in Statistical Independence and Bayesian Network Scoring Functions. Preprints2022, 2022020254. https://doi.org/10.20944/preprints202202.0254.v1
APA Style
Gogoshin, G., & Rodin, A. (2022). Resolution Limit in Statistical Independence and Bayesian Network Scoring Functions. Preprints. https://doi.org/10.20944/preprints202202.0254.v1
Chicago/Turabian Style
Gogoshin, G. and Andrei Rodin. 2022 "Resolution Limit in Statistical Independence and Bayesian Network Scoring Functions" Preprints. https://doi.org/10.20944/preprints202202.0254.v1
Abstract
In this paper we consider the congruence problem that arises in the post-analysis of Bayesian network models reconstructed from different datasets. Apart from the structure, a typical network numerically encodes relationship intensities, assigning numerical score to network edges via the scoring criterion used in the reconstruction process. This scoring is rarely a directly interpretable quantity with proper units of measure and an absolute scale, and often comes short in desirable characteristics of a true metric. This leads to poor portability of edge magnitude considerations between similar networks, originating from different sources. In this work, we address this problem by estimating the effect that data-specific resolution limit has on conditional independence, as reflected by information-theoretic entropy, and by the appropriate modification of MDL score, which removes the inconsistency between the score components in both the meaning and units. We also numerically validate our findings and expose additional performance advantages obtained via this modification.
Computer Science and Mathematics, Applied Mathematics
Copyright:
This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
