Adaptive Graphical Pattern Recognition Beyond Connectionist-Based Approaches

摘要

This paper proposes a general framework for the development of a novel approach to pattern recognition which is strongly based on graphical data types. These data keep at the same time the highly structured representation of classical syntactic and structural approaches and the subsymbolic capabilities of decision-theoretic approaches, typical of connectionist and statistical models. Like for decision-theoretic models, the recognition ability is mainly gained on the basis of learning from examples, that, however, are strongly structured.