The Machine Learning group at the Dipartimento di Informatica, Università di Torino is part of a larger research group on Artificial Intelligence, and is active since 1984. In 1998, due to the birth of the Università del Piemonte Orientale, part of the group joined the new University and started a fresh research group there.

Research Interests

The research in ML has been focussed, at the beginning, on the development of a symbolic concept learning system, ML-SMART, able to acquire a network of first-order logic classification rules in noisy domains. The system, initially based on inductive techniques, has been extended, later, with a deductive component and, more recently, also by including abductive reasoning, the ability of learning intensional definitions of relations and of handling continuous-valued numerical features (SMART+). With an incremental version of this system, ENIGMA, an industrial diagnostic expert system has been learned and used successfully in field. Another system, RIGEL, also devoted to the acquisition of first-order concept descriptions, has been developed along the same research line. New directions in symbolic ML have been explored. On one hand, an abductive reasoning mechanism, using deep models of the domain, has been implemented in a new system WHY. This system served also as a tool to explore the possibility of speeding up learning by suitably selecting the examples, according to the deep model, and their presentation order. On the other hand, issues in knowledge representation for ML, such as exploitation of abstraction theories, have been proposed. More recently, the research themes have been extended to also include Genetic Algorithms and Neural Networks. In particular, the system REGAL, able to learn first-order logic concept descriptions using a parallel genetic search, has been designed and implemented. A new selection operator has been proposed and a theoretical analysis of its behaviour has been performed. On the neural network side, the focus of attention is about locally receptive fields networks (Radial Basis Functions and Fuzzy Logic Controllers), in particular about how symbolic knowledge, generated with classical ML approaches, can be used to initialize a network, and how a network can be refined by means of Reinforcement Learning. Finally, also some work has been done in COLT, in learning recursive concept definitions, and in semi-automated knowledge elicitation methods. The target of another currently in progress research project, consists in using Mobile Agents for automatic information retrieval, network management and security purposes within a distributed system. A fundamental aspect of this architecture is the capability of using ILP methods in order to improve the agents functionalities. The group receives funds from the Italian Education and Research Departments, from the National Research Council (CNR) and from several cooperations with industries. Moreover, the group has been involved in ESPRIT Projects, namely on Special Algorithms and Architectures for Speech and Image Processing, and on Learning in Robotics.