Speaker: Alessio Martino, La Sapienza Università di Roma
Graphs are powerful mathematical entities able to capture topological and semantic information from the data at hand. Not by chance they have been widely used to model several complex systems in a plethora of domains (including, but not limited to, biology, social networks, computer vision). The widespread use of such fascinating structures intrigued machine learning engineers and computer scientists alike for more than two decades. Yet graphs, by definition, are able to capture only pairwise relationships amongst vertices and this can limit their modelling power. Hypergraphs fill this gap by allowing hyperedges to connect simultaneously two or more vertices together. In this talk, I will be presenting my latest research in the development of advanced pattern recognition techniques in the graph domain and in the hypergraph domain for solving mainly supervised learning problems. The presented techniques span several approaches, including kernel methods, embedding techniques and feature engineering. In order to demonstrate their effectiveness, two real-world biological case studies will also be discussed, alongside common benchmark tests.