home · mobile · calendar · colloquia · 2010-2011 · 

Colloquium - Mozer

Study Less, Learn More: Optimizing Retention Using Computational Models of Human Memory
Department of Computer Science

In school and beyond, individuals need to learn facts: proficiency in a foreign language involves knowing the translation of English words; becoming a physician requires associating symptoms with diseases; surviving in the wilderness may depend on identifying whether a snake or berry is poisonous. We are developing techniques to assist individuals in learning and retaining knowledge. These techniques are based on computational cognitive models of memory, which predict recall accuracy under differing training conditions (e.g., duration, frequency, and spacing of study). Using queries to the model, Bayesian integration over uncertainty, and numerical optimization, we can determine the training conditions that are expected to obtain the most durable memories. We describe two cognitive models, one that predicts retention as a function of the temporal lags between study episodes, and another that predicts which individual item in a set will most benefit from additional study. We have validated these models using behavioral experiments, and we are currently developing web-based tutors that utilize model predictions to optimize retention of facts.

This work is a collaboration with Robert Lindsey and Owen Lewis (University of Colorado), Harold (Hal) Pashler, Sean Kang, and Edward Vul (UCSD), and Nicholas Cepeda (York University).

Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:13)