Hybrid POMDP for Mobile ITS with Co-optimized Pedagogical Effectiveness and Mobile Performance
MSc Thesis Defense by:
Vimanga Umange
Date: August 28th 2026
Time: 1-3 PM
Location: MS Teams
Abstract:
Mobile Intelligent Tutoring Systems (ITS) must provide personalized and engaging instruction while operating under the latency, energy, and hardware constraints of mobile devices. This thesis proposes a hybrid Partially Observable Markov Decision Process (POMDP) framework that jointly considers pedagogical effectiveness and mobile performance. The framework uses Knowledge Space Theory to model student mastery and a Deep Q-Network to select, at each tutoring step, what to teach, where computation should be executed across local, edge, or cloud resources, and which interface modality should be presented. A multi-objective reward function balances learning gain, latency, energy consumption, and modality appropriateness. Experimental evaluation across multiple device classes, network conditions, and trade-off configurations shows that the learned policy can substantially improve learning performance while reducing response latency and maintaining real-time operation across most tutoring interactions. The results demonstrate the potential of reinforcement learning to provide unified, adaptive control of both pedagogical and technical decisions in mobile ITS.
Keywords: Intelligent Tutoring Systems (ITS), Partially Observable Markov Decision Process (POMDP), Deep Reinforcement Learning, Mobile Edge Computing, Knowledge Space Theory
Thesis Committee:
Internal Reader: Dr. Jessica Chen
Internal Reader 2: Dr. Muhammad Asaduzzaman
Advisor: Dr. Xiaobu Yuan
Chair: Dr. Peter Tsin
This is a closed defense and not open to the public
