Home // INNOV 2025, The Fourteenth International Conference on Communications, Computation, Networks and Technologies // View article

A Note on Structure Compatibility for Large Scale Structure Learning

Authors:
Sung-Ho Kim
Namgil Lee

Keywords: combined model structure; Markovian subgraph; structural discrepancy; union graph

Abstract:
Suppose that we are given two statistical model structures given in graphs. We are interested in testing whether they are from one source model or data. If the models share a source, we say that the models are compatible. In the paper, we present methods of testing compatibility of two model structures provided that the two structures share at least two nodes. The model structure represents causal or associative relationships between random variables (or nodes in graphs). Two testing methods will be proposed. One is by comparing structures of the intersection part of the two models, and the other is by using what we call union graphs. A union graph is obtained by merging the given structures with some additions and deletions of edges under a specified condition. We then check if the given structures are possible from the union graph. The methods are illustrated through examples. We aim to develop a method of structure learning by using as many pieces of structure information as possible. In this line of work, the pieces of information given in graphs need be checked for compatibility among themselves. This is the reason why this small piece of work is so crucial to the success of our future work.

Pages: 1 to 4

Copyright: Copyright (c) IARIA, 2025

Publication date: September 28, 2025

Published in: conference

ISSN: 2326-9286

ISBN: 978-1-68558-302-6

Location: Lisbon, Portugal

Dates: from September 28, 2025 to October 2, 2025