Optimal Learning Lab

Applying cognitive theory to optimize learning.
Menu
Research

Building learning systems that can justify their choices.

The lab's research program centers on theories of learning that are explicit enough to guide software decisions. We study how memory, prior practice, item relationships, and learner performance can be modeled in ways that improve instructional timing, sequencing, and support.

Themes

Core lines of inquiry

Memory and Practice Scheduling

How should practice be spaced, revisited, and sequenced so that performance later is stronger and more durable?

Knowledge Tracing and Prediction

How can learner histories be translated into interpretable models that predict performance and support instructional decisions?

Learning With Connected Knowledge

How should adaptive systems handle material with dependencies across concepts, propositions, and mental models instead of isolated facts alone?

Approach

From formal theory to instructional software

Model the learner

We develop models that use prior practice history, timing, success, failure, and content structure to estimate how learning changes over time.

Use models to decide

We use those models to choose what a learner should practice next, when review should happen, and how practice should be organized.

Test in realistic settings

We evaluate whether model-based decisions improve learning in applied educational contexts, not just in abstract simulations.

Build tools others can use

MoFaCTS offers adaptive practice, LKT provides reusable learner modeling in R, and TutorScript/SPARC investigates how explicit instructional rules can guide generative AI. Together they connect theories of learning to tools for research and application.

Research Questions

Representative questions

Optimization

When is it worth practicing one item instead of another?

We study how to compare competing practice opportunities so educational systems can allocate limited time more effectively.

Interpretation

What kinds of learner data lead to useful, interpretable models?

We examine feature design, knowledge tracing structure, and the tradeoff between predictive accuracy and explanatory clarity.

Transfer

How should systems adapt when knowledge is interconnected?

Many domains involve dependencies among concepts. The lab studies how those dependencies should alter practice decisions and model structure.

Working principle

A learning system should be able to explain why a practice decision is appropriate for this learner, at this moment, for this content.
TutorScript/SPARC investigates this requirement by separating instructional decisions from AI-generated language: learner models provide evidence, rules select eligible actions, and generative AI produces the selected content. Researchers can inspect those choices and compare instructional policies. Explore the projects or read the TutorScript guide.