One of the goals of computerized tutoring systems is to optimize the learning of facts. Over a hundred years of declarative memory research have identified two robust effects that can improve such systems: the spacing and the testing effect. By making optimal use of both and adjusting the system to the individual learner using cognitive models based on declarative memory theories, such systems consistently outperform traditional methods. This adjustment process is driven by a continuously updated estimate of the rate of (...) forgetting for each item and learner on the basis of the learner's accuracy and response time. In this study, we investigated to what extent these estimates of individual rates of forgetting are stable over time and across different materials. We demonstrate that they are stable over time but not across materials. Even though most theories of human declarative memory assume a single underlying rate of forgetting, we show that, in practice, it makes sense to assume different materials are forgotten at different rates. If a computerized, adaptive fact-learning system allowed different rates of forgetting for different materials, it could adapt to individual learners more readily. (shrink)

One of the goals of computerized tutoring systems is to optimize the learning of facts. Over a hundred years of declarative memory research have identified two robust effects that can improve such systems: the spacing and the testing effect. By making optimal use of both and adjusting the system to the individual learner using cognitive models based on declarative memory theories, such systems consistently outperform traditional methods (Van Rijn, Van Maanen, & Van Woudenberg, 2009). This adjustment process is driven by (...) a continuously updated estimate of the rate of forgetting for each item and learner on the basis of the learner's accuracy and response time. In this study, we investigated to what extent these estimates of individual rates of forgetting are stable over time and across different materials. We demonstrate that they are stable over time but not across materials. Even though most theories of human declarative memory assume a single underlying rate of forgetting, we show that, in practice, it makes sense to assume different materials are forgotten at different rates. If a computerized, adaptive fact‐learning system allowed different rates of forgetting for different materials, it could adapt to individual learners more readily. (shrink)

The fields of machine learning (ML) and cognitive science have developed complementary approaches to computationally modeling human behavior. ML's primary concern is maximizing prediction accuracy; cognitive science's primary concern is explaining the underlying mechanisms. Cross-talk between these disciplines is limited, likely because the tasks and goals usually differ. The domain of e-learning and knowledge acquisition constitutes a fruitful intersection for the two fields’ methodologies to be integrated because accurately tracking learning and forgetting over time and predicting future performance based on (...) learning histories are central to developing effective, personalized learning tools. Here, we show how a state-of-the-art ML model can be enhanced by incorporating insights from a cognitive model of human memory. This was done by exploiting the predictive performance equation's (PPE) narrow but highly specialized domain knowledge with regard to the temporal dynamics of learning and forgetting. Specifically, the PPE was used to engineer timing-related input features for a gradient-boosted decision trees (GBDT) model. The resulting PPE-enhanced GBDT outperformed the default GBDT, especially under conditions in which limited data were available for training. Results suggest that integrating cognitive and ML models could be particularly productive if the available data are too high-dimensional to be explained by a cognitive model but not sufficiently large to effectively train a modern ML algorithm. Here, the cognitive model's insights pertaining to only one aspect of the data were enough to jump-start the ML model's ability to make predictions—a finding that holds promise for future explorations. (shrink)

The fields of machine learning (ML) and cognitive science have developed complementary approaches to computationally modeling human behavior. ML's primary concern is maximizing prediction accuracy; cognitive science's primary concern is explaining the underlying mechanisms. Cross-talk between these disciplines is limited, likely because the tasks and goals usually differ. The domain of e-learning and knowledge acquisition constitutes a fruitful intersection for the two fields’ methodologies to be integrated because accurately tracking learning and forgetting over time and predicting future performance based on (...) learning histories are central to developing effective, personalized learning tools. Here, we show how a state-of-the-art ML model can be enhanced by incorporating insights from a cognitive model of human memory. This was done by exploiting the predictive performance equation's (PPE) narrow but highly specialized domain knowledge with regard to the temporal dynamics of learning and forgetting. Specifically, the PPE was used to engineer timing-related input features for a gradient-boosted decision trees (GBDT) model. The resulting PPE-enhanced GBDT outperformed the default GBDT, especially under conditions in which limited data were available for training. Results suggest that integrating cognitive and ML models could be particularly productive if the available data are too high-dimensional to be explained by a cognitive model but not sufficiently large to effectively train a modern ML algorithm. Here, the cognitive model's insights pertaining to only one aspect of the data were enough to jump-start the ML model's ability to make predictions—a finding that holds promise for future explorations. (shrink)

The parameters governing our behavior are in constant flux. Accurately capturing these dynamics in cognitive models poses a challenge to modelers. Here, we demonstrate a mapping of ACT-R's declarative memory onto the linear ballistic accumulator (LBA), a mathematical model describing a competition between evidence accumulation processes. We show that this mapping provides a method for inferring individual ACT-R parameters without requiring the modeler to build and fit an entire ACT-R model. Existing parameter estimation methods for the LBA can be used, (...) instead of the computationally expensive parameter sweeps that are traditionally done. We conduct a parameter recovery study to confirm that the LBA can recover ACT-R parameters from simulated data. Then, as a proof of concept, we use the LBA to estimate ACT-R parameters from an empirical dataset. The resulting parameter estimates provide a cognitively meaningful explanation for observed differences in behavior over time and between individuals. In addition, we find that the mapping between ACT-R and LBA lends a more concrete interpretation to ACT-R's latency factor parameter, namely as a measure of response caution. This work contributes to a growing movement towards integrating formal modeling approaches in cognitive science. (shrink)

The parameters governing our behavior are in constant flux, and capturing these dynamics in cognitive models remains a challenge. We demonstrate how a mapping between ACT‐R's model of declarative memory and the linear ballistic accumulator enables efficient estimation of memory parameters from data. The resulting estimates provide a cognitively meaningful explanation for observed differences in behavior over time and between individuals.