An overview of the projects I am working on.
My main interest is in finding ways to adapt fact-learning to individual learners and thereby yield better learning outcomes. Throughout my PhD project, I have worked with Hedderik van Rijn on the system we call SlimStampen (all projects mentioned in this category are in collaboration with Hedderik). The model uses input from the learner recorded during learning to continuously update item-level parameters that indicate how quickly this learner is expected to forget the item being studied. It uses these estimates to make sure items are repeated before they are forgotten.
A lot of the work in this context focuses on investigating these estimated parameters in more detail. We looked at the stability of the parameter estimates over time and across different materials and found that the parameters could be estimated reliably over a three-week period. The results from this project were published in Topics in Cognitive Science.
We also explored how those estimated parameters are related to established measures of individual differences: working memory capacity and general cognitive capacity. Preliminary results indicated that there is no correlation between our model’s parameters and those two measures – we reported these findings in the CogSci conference proceedings and presented results at the annual meeting in August 2016. A manuscript discussing the full dataset has been submitted in May 2018 (also see Chapter 4 in my PhD thesis).
Another interesting question is to which extent motivation plays a role when studying with our adaptive method. To learn more about this aspect, we pre-registered an experiment to test the effect of different levels of rewards on the model’s parameters. The details about this project are documented on the Open Science Framework. We completed data collection at the end of the academic year 2015/16. Preliminary results of this project were presented as a poster at Psychonomics 2016 [blog post; link to poster on the OSF]. This work is conducted in collaboration with Berry van den Berg and Don van Ravenzwaaij.
On top of that, we are currently revising a manuscript from a study in which we compared our adaptive learning method to a decently effective flashcard “control” condition in a within-subject design. Previous work suggests that performance on a subsequent test is better when studying with the adaptive method but comparing the methods in a within-subject design will hopefully allow us to say more about who benefits the most from studying with a personalized learning system.
Together with Maarten van der Velde and Jelmer Borst, we are now starting to mine big datasets to explore ways to improve the adaptive system. A first exploration of this line of work will be presented at ICCM 2018 in Madison in July.
CPR as a Model for Learning Complex Skills
An alternative line of research that we (Hedderik and I, together with Sarah Maass) embarked on is that of learning complex skills that require more than declarative knowledge. We received a grant from the European Office of Aerospace Research and Development to learn more about how declarative and procedural learning interact when learning a complex skill. The complex skill we chose to investigate is cardiopulmonary resuscitation (CPR) which is part of first aid. This project is a collaboration with Kevin Gluck’s group at the Air Force Research Laboratory. We showcased initial findings at ICCM 2016 (poster as PDF) and ICCM 2017.
Data collection for a large study with retention intervals of multiple months was completed by Sarah in late May of 2018 and we are now digging into the data to prepare a manuscript with our main findings. This work is conducted with collaborators from the Air Force Research Laboratory (AFRL), specifically Micheal Krusmark and Kevin Gluck (with help from Matt Walsh and Siera Martinez).