Kernel-Based Semi-Supervised Learning Over Multilayer Graphs

Networks arise in fields such as sociology, biology, and machine learning among others, to describe complex and often interdependent systems. These increasingly complex systems call for flexible network models that allow for multiple types of interactions among the agents (nodes) known as multilayer networks. A frequently encountered task entails inference of nodal processes across the network given values on a subset of nodes. The present contribution relies on graph kernels, to put forth a novel inference approach that accounts for linear and nonlinear dependencies among nodes and leverages the layered network structure. Numerical tests with synthetic as well as real data corroborate the effectiveness of the proposed kernel-based multilayer learning scheme.