Artificial Intelligence

Artificial Intelligence: Uncertain Temporal Reasoning in the frameworks of Constraint Programming and Fuzzy Logics

The first line of research concerned Artificial Intelligence, and namely Uncertain Temporal Reasoning in the frameworks of Constraint Programming and Fuzzy Logics. The main aim was the development of advanced tools for representing vague and uncertain temporal information and reason about it. Two prototypes were implemented: a first one in C++, to efficiently handle hundreds of constraints, and a second one in Prolog Constraint Handling Rules (a Logic Programming framework) to explore more flexible solutions, such as “coarse” information and Spatial Reasoning. Potential applications of these systems, besides the traditional field of schedulers, were devised in Diagnostics and in Computational Genomics. The novelty of the developed systems was the management of integrated fuzzy qualitative-metric constraints; the uncertainty was managed by providing preference degrees to the variables and to the constraints that define the problems, in order to automatically find the redundant constraints or the partially inconsistent constraints, and optimize the solutions according to their resulting global preferences. This behavior is similar to the human reasoning, which is often able to define more easily what is more plausible instead of what is more probable. Since the theoretical models of the solvers had an exponential complexity in the general case, Dr. Marco Falda studied the tractability and the complexity of the qualitative algebras underlying them. At the Department of Pure and Applied Mathematics he studied Conditional Temporal Problems with Preferences under the supervision of Professor F. Rossi; during this period he was invited to present a tutorial entitled “Fuzzy Temporal Reasoning” at the International Conference on Advanced Planning and Scheduling (ICAPS’07), Brown University, Providence (USA).

As far as Diagnostics applications are concerned, temporal constraints were applied to model phenomena that progress over time following typical trends. A first application of the constraint solver developed in this area was about the diagnosis of exanthematic diseases from their typical evolution and from inaccurate reports of patients: a temporal model was created for every known disease (5 in the cited paper) and for the symptoms elicited from the patients; then a temporal consistency analysis was carried out to identify the disease most plausible. Reversing the problem, in the temporal evolution of the symptoms that characterize a little known disease was described, modeling it as a set constraints; then the consistency was checked with respect to the patients reports, inferring the most common temporal features.

Comments are closed