ML-SMART: A Problem Solver for Learning from Examples

Article type: Research Article

Authors: Bergadano, F.^a | Gemello, R.^b | Giordana, A.^a | Saitta, L.^a

Affiliations: [a] Dipartimento di Informatica, Università di Torino, C.so Svizzera 185, 10149 Torino, Italy | [b] C.S.E.L.T., Via Reiss Romoli 274, 10148 Torino, Italy

Abstract: This paper describes ML-SMART, a knowledge based system for inducing conceptual description from examples. It is organized as a problem solver and makes an integrated use of multiple search strategies, including characterization, constructive learning and deduction. The acquired knowledge consists of production rules organized into a network, which can be seen as a generalization of a decision tree. The condition part of the rules is expressed in a first order logic language also containing numerical quantification. The use of variables and functions allows highly structured concepts to be easily described. ML-SMART (Similarity-based Multiple-concept Acquisition and Reasoning Tool) is a domain independent system, provided with a friendly interface to ease the task of configuring a new application. It is well suited to work in noisy environments and has been tested in different domains for the sake of comparison with other existing systems.

DOI: 10.3233/FI-1989-12104

Journal: Fundamenta Informaticae, vol. 12, no. 1, pp. 29-49, 1989