Active Learning Map: A Novel Assimilation Guided Adaptive Learning Platform

Abstract

Identifying individualized assimilation patterns of students can play an important role in personalizing learning interventions and improving their learning outcomes. While studying from similar course materials, learners frequently exhibit significant variability in how they internalize and prioritize core concepts. Recognizing this variability helps in designing meaningful personalized interventions, which in turn can lead to greater learner engagement and outcomes. An effective way to represent assimilation patterns is in the form of a high dimensional data structure called a Polyline that captures the differentiated way in which a given student has assimilated knowledge across different topics in the course. In this work, we extend this representational concept to create a model free Deep Reinforcement Learning Framework that can provide personalized recommendations to learners based on their Polylines. This is then visualized on a two-dimensional progression space called an Active Learning Map that enables the assimilation patterns to be visually displayed and interpreted. We cluster learners based on assimilation pattern and learning style using a Gaussian Mixture Model and train a Conditional Generative Adversarial Network (CGAN) as an Assimilation Estimator to simulate how their Polylines evolve. Using this CGAN as an environment, we train a Deep Q-Network (DQN) to discover individualized recommendation policies. Simulations demonstrate that this CGAN based environment enables the DQN to find highly effective policies using a very small sample of real student data. Furthermore, we find our DQN recommendation engine consistently outperforms a random policy, a coherent resource-matching heuristic, and a Proximal Policy Optimization (PPO) algorithm. Our proposed system functions as an autonomous intelligent agent within the Learning Map and acts as a guide, helping learners to visualize exactly where they are among a complex web of topics and lessons, as well as providing the best resources to help the learners close their knowledge gap.