Edmondo Trentin (University of Siena)

Dec 18, 2019 – 11:00 AM

DIISM, Artificial Intelligence laboratory (room 201), Siena SI

Description

Structured data in the form of labeled graphs (with variable order and topology) may be thought of as the outcomes of a random graph generating process, characterized by an underlying probabilistic law. In this talk I will first review the notions of generalized random graph (GRG) and of probability density function (pdf) for GRGs. Then, I will summarize the main facets of a “universal” learning machine for estimating the unknown pdf underlying an unsupervised sample of GRGs. The machine relies on a maximum likelihood training algorithm, constrained so as to ensure that the resulting model satisfies the axioms of probability. Techniques for preventing the model from degenerate solutions are illustrated, as well as variants of the algorithm suitable to the tasks of graphs classification
and graphs clustering. Finally, I will outline an on-going research activity aimed at exploiting such an estimator within the context of a
probabilistic graphical model suitable for learning and decision-making over sequences of GRGs.