'COMPO' conceptual clustering with connectionist competitive learning
Hugo de GARIS · Dépôt institutionnel de l'Université libre de Bruxelles (Université Libre de Bruxelles) · 1989
This paper introduces the idea that conceptual clustering can be performed using connectionist competitive learning. Competitive learning is used to detect clusters of objects and their corresponding (qualitative) descriptions. A genetic algorithm is employed to choose a subset of these descriptions such that the objects matching them form partitions over the population of objects concerned. Hierarchical classification trees are built by recursing the above two steps (competitive learning 'clustering' and genetic algorithm 'partitioning') over the objects matching the descriptions at each node.