Adaptive scheduling systems aim to estimate the ability of an individual in order to prescribe a personalized training schedule. These adaptive systems are often founded on regularities of human memory such as a learning, forgetting, and the spacing effect. One such model which has been developed to both account for regularities of memory and be used in applied contexts is the Predictive Performance Equation (PPE). One limitation of the PPE is that it is only able to account for and incorporate information about a participant’s accuracy on a task and cannot take into account additional performance measures such as reaction time. To expand the PPE, we propose a simple extension to the model, allowing it to account for both accuracy and reaction time measures. Our paper reports the extension to the PPE as well as a formal model comparison to another model of learning and retention (Pavlik and Anderson, 2005). The results of our model comparison reveal that the extended PPE can both better account and predict an individual’s performance than Pavlik and Anderson (2005) model.

In applied settings, computational models of memory haveproven useful in making principled performance predictions.Specifically, historical data are used to derive modelparameters in order to enable out-of-sample predictions.Parameters are typically fit to meaningful subsets of data.However, labels that demarcate what constitutes a“meaningful” subset are not always available. Here, we utilizea data-driven method to cluster past performance into subsetspossessing statistical similarities. We contrast predictions fromcluster-specific model parameters with predictions based onsubsets that are artifacts of the experimental design. We showthat cluster-based predictions are at least as accurate as thechosen baselines and highlight additional advantages of thedata-driven approach.

Human learning has been characterized by three robust effects (i.e.power law of learning, power law of decay, and spacing), whichhave been validated across multiple domains and time intervals. Toaccount for these different effects mathematical model of learningand retention have been developed. These models hold a great dealof potential for application a wide range of educational and trainingscenarios. However, many models are not validated according fortheir ability to make accurate predictions of human performance.The predictive demand of these models is made increasinglycomplex by the needs of training domain, needing both to predictboth skill decay and reacquisition from little historical data. In thispaper, we examine the predictive capability of the PredictivePerformance Equation (PPE) implemented in a Bayesianhierarchical model. Through a comparison of two Bayesianhierarchical models we show how hierarchical model fit to aparticipant’s performance across a set of items compared to only asingle item improves PPE’s predictive accuracy of both skill decayand reacquisition over multiple learning schedules

Retrieval practice—the process of actively calling information to mind rather than passively studying materials—has been proven to be a highly effective learning strategy. However, only recently, researchers have started to examine differences between learners in terms of the optimal conditions of retrieval practice in applied educational settings. In this study (N = 118), we focus on learners with dyslexia: a group that is usually not included in the retrieval practice literature. We compare their performance to the performance of typical learners in an adaptive, retrieval practice-based word learning task using both typing-based and speech-based response conditions. We find that typical learners outperform learners with dyslexia when they are asked to respond by typing, but that this difference is much smaller when learners can respond by speech. These results can contribute to the development of educational technology that allows for effective and inclusive learning in neurodivergent individuals.

Adaptive fact learning systems have been developed to makeoptimal use of testing and spacing effects by taking intoaccount individual differences in learning efficiency.Measures derived from these systems, capturing theindividual differences, predict later performance in similarand different fact learning tasks. Additionally, there is a richbody of literature showing that individual differences ingeneral cognitive ability or working memory capacity canpredict scores on achievement tests. If these measures alsoinfluence fact learning, incorporating them might furtherenhance adaptive systems. However, here we provideevidence that performance during fact learning is neitherrelated to working memory capacity nor general cognitiveability. This means that the individual differences captured byour adaptive learning system encapsulate characteristics oflearners that are independent of their general cognitive ability.Consequently, adaptive learning methods should focusprimarily on memory-related processes.

Digital adaptive fact learning systems tailor learning sessions to the individual learner. Most adaptive learning systems assume that the learner has no prior knowledge of the material and therefore use an initial ‘passive rehearsal’ trial for all facts. Here, we test a system that uses active retrieval trials instead of passive study trials when introducing items. This minimizes time loss associated with studying familiar materials and maximizes potential benefits of attempted retrieval before study. We test the system by having participants learn the association between the outlines and names of countries, a domain in which they are likely to have varying degrees of prior knowledge. We show that using attempted retrieval to identify known items is valuable in real-world applications, where partial knowledge of study materials before the start of a learning session is very common.

Models of memory that rely on a single forgetting curve can struggle to simultaneously capture memory performance over short and long timescales. Typically, model parameters that account well for forgetting over a span of seconds or minutes are too pessimistic when it comes to predicting longer-term retention over days or weeks. Crucially, it is precisely the ability to model memory performance within and between study sessions that is valuable in applied educational settings. Previous attempts to address the underestimation of long-term retention involved scaling down between-session time. Here, we investigate how the optimal scaling factor changes as a function of the interval between study attempts, using a rich naturalistic dataset with intervals ranging from several seconds to multiple weeks. By fitting an ACT-R model, we show that the optimal scaling factor changes with time and interacts with the retrieval threshold. This finding has direct practical and theoretical implications.

Cardiopulmonary resuscitation (CPR) is the single greatest determinant in survival from cardiac arrest and is an essential part of every medical professional’s toolkit. Effectively responding to cardiac arrests requires CPR proficiency, which is best retained through frequent training. One such model is Resuscitation Quality Improvement (RQI) that requires users to exhibit proficient performance in compression and ventilation skills every 3 months. While frequent refresher trainings are superior to the traditional 24-month intervals (Cheng et al., 2020), they still need to consider inter-individual differences. Healthcare environments are rife for personalized, adaptive approaches to tracing users’ proficiency over time. To this end, we explore how users that perform similarly over time can be clustered together. Such performance profiles could ultimately enhance the benefits of frequent training with personalized efficiency for users with different needs. With the long-term goal of building an adaptive scheduling tool in mind, we present some initial explorations in this domain. Using k-means clustering, we show that a small number of clusters seems sufficient to create meaningful performance profiles to make out-of-sample predictions. Furthermore, our simulation study suggests that fuzzy membership to said clusters can be leveraged to enhance predictions. We discuss potential next steps in which these fuzzy performance profiles can be employed by more powerful predictive models to move the field towards personalized, adaptive training schedules that improve learning efficiency with the goal of increasing survival outcomes after cardiac arrest.

A complete and holistic understanding of human cognitive function using a cognitive model should be able to incorporate both the idiographic biological parameters from behavioral data and interactions between connected brain networks identified by neuroanatomical techniques. Here, we argue that first, computational modeling can be used to extract biological parameters, and second, the biological parameter should be identifiable through the corresponding brain functional networks. We tested this empirically with the long-term memory’s decay rate (rate of forgetting α) measured from a Swahili-English vocabulary learning task, and resting-state fMRI (rs-fMRI) data, both collected from 33 participants. α was estimated based on a rational model of episodic memory. rs-fMRI data from each brain was processed into a 264 x 264 connectivity matrix, and further selected and grouped into memory-related network matrices using group lasso and cross-validation techniques. We were able to (1) predict the observed α using these connectivity matrices, and (2) show that forgetting can be related to specific brain networks.

Spacing is a highly effective encoding strategy that has been shown to benefit memory in a variety of domains. Recent work has emphasized the evaluation of spaced practice under conditions that more closely reflect daily life. This work found that spacing repetitions over days, weeks, or months is effective over retention intervals as long as one year. One aspect of spaced study that has received less attention, however, is the relationship between the number of acquisition sessions and final retention. That is, if a student preparing for an exam plans to allocate 10 hours to preparing for that exam, is there an optimal, or perhaps minimal, number of study sessions that they should engage in to best leverage the benefits of spacing? In the present experiment we had participants complete 16 practice tests of Japanese-English pairs (e.g., boushi – hat). These practice tests were either all completed in one session, or distributed across two, three, or four sessions. These sessions were spaced either 1, 7, 30, or 90 days apart. Participants completed four test trials following a retention interval of 90 days, 180 days, or some variable length. Our results suggest that the number of acquisition sessions monotonically enhanced first-trial test performance as well as relearning, though evidence for enhanced relearning between one and two sessions was ambiguous. Unexpectedly, these monotonic trends were stable across practice lags and retention intervals. These findings suggest that, in addition to the temporal lag between practice episodes, the number of sessions over which one elects to distribute those episodes also has ramifications for long-term retention, and that each additional session yields meaningful benefits.