Electronic Theses and Dissertations

Kevin Russell

Date

2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Mathematical Sciences

Committee Chair

Alistair Windsor

Committee Member

Ching-Chi Yang

Committee Member

Majid Noroozi

Committee Member

Philip Pavlik

Abstract

Personalized tutors often learn parameters in batch and update them only intermittently, which limits adaptation when a learner’s state changes. Within Logistic Knowledge Tracing (LKT), recent work addresses cold start issues by engineering temporal features that encode recency and history in the inputs (e.g., Pavlik and L.G. Eglington (2025)). This dissertation advances LKT from feature-level to procedural adaptivity by introducing true online optimization for logistic models and evaluating its effect on real-time individualization. I instantiate online LKT with three adaptive gradient methods (Adagrad, RMSProp, Adam) that update coefficients after each interaction. A time-respecting walk-forward protocol anchors all comparisons: models are trained on earlier portions of each sequence and evaluated on strictly future data. Experiments on three benchmark datasets spanning distinct domains and temporal structures (Cloze completion, fraction arithmetic with blocked versus interleaved practice, and Mathia algebra) compare online LKT to matched batch LKT baselines. BKT is included for reference. Results show that online LKT reaches near-peak performance with very limited evidence, consistently matching or exceeding the best batch variants in the early-data regime (1–5% of the stream), and remains competitive as more data accrue. Differences among Adagrad, RMSProp, and Adam are modest overall, with adaptive per-parameter learning rates typically offering the fastest rise to strong performance when data are sparse. Methodologically, the work reframes LKT as a continuously adapting learner model rather than a static regression augmented by temporal features. Practically, it offers a simple, interpretable path to real-time individualization, which we evaluate under causal, time-ordered conditions.

Comments

Data is provided by the student.”

Library Comment

Dissertation or thesis originally submitted to ProQuest.

Notes

Embargoed until 11-17-2026

Download

Available for download on Tuesday, November 17, 2026

Share

COinS
 

Archival Statement

This item was created or digitized prior to April 24, 2026, or is a reproduction of legacy media created before that date. It is preserved in its original, unmodified state specifically for research, reference, or historical recordkeeping. This material is part of a digital archival collection and is not utilized for current University instruction, programs, or active public communication. In accordance with the ADA Title II Final Rule, the University Libraries provides accessible versions of archival materials upon request. To request an accommodation for this item, please submit an accessibility request form.