Relational models of feature based concept formation, theory-based concept formation and analogical retrieval/mapping

TL;DRAbstract

This thesis draws together a number of areas of cognitive computational research including symbolic connectionism and the modelling of concept formation and analogy. Central to the work is the processing of relations in a hybrid neural symbolic architecture. Models of analogy have recognised the importance of relational representations for some time now, and psychological research into concept formation has investigated the role of relational knowledge. However computational models of concept formation have typically not incorporated this into their architectures. This thesis formalises a tensor representation of relations and develops two models based on this formulation. One model is the Relcon model that applies an iterative procedure to a tensor representation of relations in order to converge to a stable representation of concepts. In addition to incorporating relational knowledge, the model accounts for typicality ratings of concept members and models a family resemblance-typica

Chat with Paper

AI Agents for this Paper