Bayesian-based models for perception and memory: ROUSE and REM

Rich Shiffrin
University of Indiana

One approach to model development is based on constrained optimality: Assuming certain limitations on cognitive processes (derived from previous research), what processing system would carry out decision making in a given task domain in such a way that a chosen performance measure would be maximized? The rational for this approach is adaptation over evolution and development. Perhaps surprisingly, some of the best cognitive models to date have had such a genesis. We discuss (time permitting) one example from perception and one from memory. The ROUSE model (2001) was developed to explain perception under difficult conditions, and the effects of short-term priming on perception. The REM model (1997) was developed to explain recognition memory phenomena and the growth of knowledge from experience.

The ROUSE model has at least three different instantiations: The original was Bayesian-inspired and both this and a generative version will be discussed. A neural implementation has also been developed, but time constraints will prevent discussion at this workshop. The ROUSE model assumes that perception derives from features extracted probabilisitically from the environment that are imprecisely located in time and space. Optimal decisions concerning what had been presented requires ‘discounting’ of certain types of evidence. This model is almost unique in cognition in predicting a priori and correctly the results of studies yet to be carried out, even when intuition and history suggest otherwise.

The REM model assumes probabilistic storage of an event in both a new memory trace and in a prior sufficiently similar trace. Its recognition memory decisions are optimal under assumptions that retrieved traces are those related to the cues used to probe memory. The resultant model predicts a surprisingly large set of results and sometimes predicts new results in quantitative fashion using the original set of parameters.

Back to Probabilistic Models of Cognition: The Mathematics of Mind