Learning Systems of Concepts with an Infinite Relational Model

Relationships between concepts account for a large proportion of semantic knowledge. We present a nonparametric Bayesian model that discovers systems of related concepts. Given data involving several sets of entities, our model discovers the kinds of entities in each set and the relations between kinds that are possible or likely. We apply our approach to four problems: clustering objects and features, learning ontologies, discovering kinship systems, and discovering structure in political data. Philosophers, psychologists and computer scientists have proposed that semantic knowledge is best understood as a system of relations. Two questions immediately arise: how can these systems be represented, and how are these representations acquired? Researchers who start with the first question often devise complex representational schemes (e.g. Minsky's (1975) classic work on frames), but explaining how these representations are learned is a challenging problem. We take the opposite appr...

• Type: Conference
• Year: 2006
• Where: AAAI
• Authors: Charles Kemp, Joshua B. Tenenbaum, Thomas L. Griffiths, Takeshi Yamada, Naonori Ueda
• Keywords: AAAI 2006, Complex Representational Schemes, Intelligent Agents, Nonparametric Bayesian Model, Semantic Knowledge