PhD Dissertation Proposal Defense: Mengxue Zhang, Scalable and Robust AI-Driven Scoring System for Open-Ended Mathematical Reasoning
Content
Speaker
Abstract
Automatic scoring student responses to open-ended mathematical reasoning questions is challenging due to their structured nature and the variability of multiple reasoning paths. This thesis explores AI-driven methods to improve the robustness and scalability of grading and feedback systems. We develop a framework that (1) analyzes student stepwise solutions to diagnose errors and provide targeted feedback, (2) introduces an improved automatic grading system leveraging domain-adapted representations and in-context learning for generalization, and (3) enhances solution coherence and controllability through a stepwise planning approach for math word problems. Collectively, these contributions enable a scalable, interpretable system for evaluating open-ended student responses across a wide range of mathematical reasoning tasks.
Advisor
Andrew Lan