AIIA 2019

18th International Conference of the Italian Association for Artificial Intelligence


Introduction to Probabilistic Graphical Models

The tutorial aims at introducing basic notions and algorithms for Probabilistic Graphical Models (PGM) in Artificial Intelligence (AI). After a short review of probability calculus and of the interpretation of probability, we will discuss different types of PGMs, namely directed models like Bayesian Belief Networks and undirected models like Markov Random Fields.

Both representational and algorithmic issues will be introduced and discussed, with particular attention to the mutual relationship between directed and undirected models. Concepts related to the conditional independence among the variables will be defined, by considering graph theoretic notion of separation in the underlying (directed or undirected) graph. Different types of inference algorithms, both exact and approximate, will be introduced, by taking into account the time requirements of the tutorial.

Examples of software tools available for the management of PGMs will be presented, by pointing out some applications in different areas.

Time: November 19, 14:00 - 18:00

Room: Aula Magna

Speaker: Luigi Portinale, University of Piemonte Orientale

Slides: PDF

Chair: Marco Calautti

Multi-agent Path Finding: from puzzles to multi-robot motion planning

This tutorial aims on introducing the problem of multi-agent path finding (MAPF) both from the theoretical perspective as well as from the point of view of practical solving algorithms and applications in robotics. The task in MAPF is to navigate agents from their initial positions to specified goal positions so that collisions between agents do not occur. We will introduce the abstract theoretical formulation of the MAPF problem that takes place in an undirected graph and will show a relation to known combinatorial puzzles like Lloyd's 15. In this regard, combinatorial solving approaches based on permutation groups will be shown. Modern-day algorithms for solving the MAPF problem optimally with respect to objectives like the total execution time or the cost will be shown too. Finally we expect to discuss the relation of the MAPF and its generalizations to motion planning and navigation in robotics.

Time: November 20, 15:00 - 19:00

Room: Aula Magna

Speaker: Pavel Surynek, Czech Technical University

Slides: PDF

Chair: Mario Alviano