Peter Gärdenfors (Lund University)
There are several way of modelling concepts. Within the symbolic tradition, logical operators have been used to construct new models. But the question of the origin of the primitive concepts is not studied. Connectionist models of concept formation can handle the creation of new concepts, but only after extensive training. Humans, on the other hand, learn new concepts very quickly. Using geometric representations in the form of conceptual spaces, I shall present an alternative model of how new concepts can be learned in an efficient way.