Solving of Minimal Mutual Information Model Problem via Regularization of Dual Problem and Using Ellipsoid Method with Inexact Oracle

Abstract

In this work, we study the problem of restoring the correspondence matrix for measuring the flows on the links of a large computer network. This is an incorrect problem, which reduces to the choice of one of the solutions of an indefinite system of linear equations. In order for the task to become correct it is necessary to define the system in which in practice there are much more correspondences than links. In this work, we consider the MMI (minimal mutual information) model, in which the problem is reformulated in the language of convex composite optimization with the quadratic residual functional and the entropy penalty.