Stochastic Learning Models

Abstract

How subjects learn has long been of interest to psychologists. Large amounts of data are available from experiments in which various outputs (responses) of subjects have been studied as functions of a range of experimenter-controlled input factors (stimuli). To account for these inputoutput relationships is a task for the theorist. In this unit we first consider a mathematical model for learning (the so-called all-or-none or one-element model) as applied by Bower [7] to a paired-associate experiment. After characterizing the formal structure of this model, we introduce a slight generalization, the two-stage all-or-none model. The analysis of a learning experiment using stimulus sampling theory and leading to a three-state Markov chain is outlined in some detail in the Exercises. These models are samples from a wide variety of stochastic learning models, mainly developed since 1950, and applicable to many different experimental situa­tions. A guide for further reading in mathematical learning theory concludes the unit.