Conditional multivariate mutual information measures

Nguyễn Quỳnh Diệp, Phạm Thọ Hoàn, Hồ Tú Bảo


Mutual information of two variables is a measure of relationship between two variables: the larger this measure the stronger the dependence, and vice visa. However, mutual information does not indicate whether the relationship between the variables is direct or indirect. To detect "direct mutual relations", we can use conditional mutual information.

In the previous studies, we have proposed the mutual information measures of multiple variables. There are many mutual information measures with more than two variables. Each of them is sensitive to a kind of relationships that may exist among the multiple variables. However, as mutual information of two variables, the multivariate mutual information measures do not show whether the multivariate relationships are direct or indirect. In this paper, we propose new multivariate conditional mutual information measures and show that they can detect indirect multivariate relationships through conditional variables.


Information theory, entropy, mutual information, biological network reconstruction.


Journal of Computer Science and Cybernetics ISSN: 1813-9663

Published by Vietnam Academy of Science and Technology