UC Riverside

An important strategy in cognitive training of working memory is to fine-tune task difficulty based on the subject's performance. In this project, we analyze data gathered from a learning experiment via video games. Our contribution is an algorithm, that optimizes game-play progression in order to overcome difficulty boundaries. To model an individual's cognitive ability growth trajectory over time and predict performance for future challenges, we integrate probabilistic graphical models with a psychometric control equation. These models are trained and tested on study datasets generated from n-back cognitive training video games: Recollect, Tapback and Recall. The Expectation Maximization (EM) technique is used, to train these models. Results confirm that, the models predicting user performance in n-levels does well and can adapt and re-estimate itself from errors in its predictions.