Computer Sciences Dept.

Learning by Pattern Induction

Stephen F. Zeigler

Human infants are able to accumulate considerable knowledge in their first year of life without having well-developed communications skills. Pattern induction is proposed as a mechanism for accomplishing learning in a precommunication environment, This mechanism acts by detecting and describing regularities in memories of past experiences. Recognizer-predictors, structures having some similarities to productions, are proposed to represent and utilize information gathered by pattern induction. In this paper, recognizer-predictor structures and the process by which pattern induction formulates them are described in the context of MUL, an existing TELOS program modelling infant-like development in a very simple environment.

Download this report (PDF)

Return to tech report index

Computer Science | UW Home