Sequence learning as Bayesian filtering

Here we present a model of sequence learning and recall based on the idea that the function of memory is to maintain an up-to-date representation of the environment that can be used to guide future perception and action. The representation of the environment is considered to be a prior which is combined with information in short-term memory to construct a posterior representation. That posterior representation drives recall and, in turn, is used to update the priors. This prediction-update cycle is a form of Bayesian filter. We apply the model to data from the Hebb (1961) task in which participants learn sequences over repeated presentations in an immediate serial recall task. The model is shown to simulate a wide range of data on the Hebb effect including the ability to learn multiple lists at once, the effect of list spacing, the differential impact of variation in the beginning versus end of lists, learning from response errors, interference between similar lists, and the effects of repetition on forward and backward recall. The model’s ability to account for these phenomena follows directly from its basic computational principles.