Prototype Learning with Attributed Relational Graphs

摘要

An algorithm for learning structural patterns given in terms of Attributed Relational Graphs (ARG's) is presented. The algorithm, based on inductive learning methodologies, produces general and coherent prototypes in terms of Generalized Attributed Relational Graphs (GARG's), which can be easily interpreted and manipulated. The learning process is defined in terms of inference operations especially devised for ARG's, as graph generalization and graph specialization, making so possible the reduction of both the computational cost and the memory requirement of the learning process. Experimental results are presented and discussed with reference to a structural method for recognizing characters extracted from ETL database.