| The amount of forgetting over time has long been known to be a negatively accelerating function, generally modeled as an exponential or power law decay function. During the forgetting process, a review session injected at the appropriate time refreshes the memory and promotes retention. Maintaining accurate memory over long periods of time, called long-term learning, is a continuous process consisting of appropriately spaced review sessions. A practical memory refresh schedule based on empirical, long-term models of memory and targeted to the individual would be an advance over today's combinations of models and systems---which are either based on just short-term empirical data or group averages.;One of the main problems in scheduling review sessions involves efficiently and effectively determining the correct sequence and timing of repetition intervals. To help address this problem, both the performance and confidence aspects of memory were explored, and distributions (histograms) of the memory measures were employed to represent the memory states. Percentage correct was used as an objective measure of the amount of forgetting, and JOLs (Judgments of Learning) were implemented as an indicator of confidence. These metacognitive JOL self-assessments can be calibrated to actual memory performance, and students can be categorized based on their self-judgment characteristics.;In addition, modeling of the storage component of memory was performed using the distributions of percentage correct and JOL as snapshots of the memory state; the changes in a distribution from one snapshot to the next denote learning or forgetting. The models in this dissertation do not include elaborative encoding, focusing instead on learning a specific set of simple independent facts. These models target the individual student and characterize the different decay aspects of his memory. Regression analysis of the decay curves resulted in the development of a 2-parameter model that can be applied to both forgetting and confidence decay.;This mathematical memory model is used as the basis for developing a review schedule for individuals. Our empirical data have demonstrated that an effective scheduling strategy can be based on expanding repetition intervals. An adaptive refresh schedule can be continually updated by using feedback from the individual's actual test performance. Thus, we have an improved long-term learning system that calculates the appropriate expanding interval sequence after characterizing the forgetting parameters specific to each student. |
