Networks represent a powerful model for problems in different scientific and
technological fields, such as neuroscience, molecular biology, biomedicine,
sociology, social network analysis, and political science. As the number of network
applications increases, so does a need for novel data analysis techniques. In many
applications, the analysis focuses on a single network to cluster or classify its nodes
or predict pairs of nodes that will form a link. In this talk, we focus on problems where
a network is a statistical unit, and the analysis regards whole networks rather than their
parts.
Methods for learning features on networks focus mainly on the neighborhood of nodes
and edges. We review some of the existing methodologies and introduce Netpro2vec,
an embedding framework based on representations of graphs based on empirical
probability distributions. The goal is to use basic node descriptions other than the
degree, such as those induced by the Transition Matrix and Node Distance
Distribution, to describe the local and global characteristics of the networks. The
framework is evaluated on synthetic and real biomedical network datasets and
compared to well-known competitors. Finally, open problems and future research
directions are highlighted.
18 Marzo 2022
Mario Rosario Guarracino
Dipartimento di Economia e Giurisprudenza, Università degli Studi di Cassino e del Lazio Meridionale, Consiglio Nazionale delle Ricerche – Italy