Ph.D XXXVIII

Supervisor: Agostino Dovier

Phone: +39 0432 558491

Room: Rizzi L2-01-ND

Mail: dreossi.talissa@spes.uniud.it

Research Project

Bridging the gap between Logic/Constraint Programming and Machine Learning

Logic Programming’s origins date back to the ’70s when researchers were debating between procedural and declarative representations of knowledge in Artificial Intelligence.

Since then, it evolved into various research branches, e.g., Constraint Logic Programming (CLP) and Answer Set Programming (ASP). These techniques are suitable for modeling real life optimisation problems, such as network routing, scheduling and so on, through logic formulas. Since the above-mentioned problems are known to be computationally intractable (they are NP-hard) one of the challenges is to find an effective way to speed-up the search for feasible (and optimal) solutions.

A possible direction is the use of Machine/Deep Learning techniques, which are already succesfully exploited in many application domains such as business/industrial-oriented settings. We can define Machine learning as the set of algorithms to automatically learn from data and (its subset) Deep Learning as the class of learning techniques that rely on the use of deep neural networks. These algorithms are able to improve many tasks, such as detecting defective products or predicting breakdowns of machinery components in industrial applications.

The research I want to pursue is, indeed, the exploitation of Machine/Deep learning techniques to boost CLP-solving. As a matter of example, one could learn new constraints or predict the best search heuristics to apply in order to speed-up the resolution of problems.