← Projects

Apprentice Learner Architecture

A computational model of learning from instruction and examples

The Apprentice Learner Architecture is a computational model of how people acquire cognitive skills from a mix of natural-language instruction and worked examples — much as a human tutor might teach a new procedure. Developed with Christopher MacLellan, Kenneth Koedinger, Daniel Weitekamp, and others, it grew out of earlier work on TRESTLE, a model of concept formation in structured domains. The architecture has been used to simulate students for testing intelligent tutoring systems, to study how the errors of simulated learners differ from those of human ones, and as the basis for “machine teaching” interfaces that let non-programmers train AI tutors by demonstration.

Work continues to build on the architecture, most recently in research on decomposed inductive procedure learning, which pushes it toward more human-like data efficiency.

Decomposed Inductive Procedure Learning: Learning Academic Tasks with Human-Like Data Efficiency
Weitekamp, MacLellan, Harpstead, Rachatasumrit, Koedinger · Cognitive Science Society, 2025
Simulating Learning from Language and Examples
Weitekamp, Rachatasumrit, Wei, Harpstead, Koedinger · AIED, 2023
Toward Stable Asymptotic Learning with Simulated Learners
Weitekamp, Harpstead, Koedinger · AIED, 2021
An Interaction Design for Machine Teaching to Develop AI Tutors
Weitekamp, Harpstead, Koedinger · CHI, 2020
Investigating Differential Error Types Between Human and Simulated Learners
Weitekamp, Ye, Rachatasumrit, Harpstead, Koedinger · AIED, 2020
A Framework for Natural Cognitive System Training Interactions
MacLellan, Harpstead, Marinier, Koedinger · Advances in Cognitive Systems, 2018
The Apprentice Learner Architecture: Closing the Loop between Learning Theory and Educational Data
MacLellan, Harpstead, Patel, Koedinger · EDM, 2016
TRESTLE: A Model of Concept Formation in Structured Domains
MacLellan, Harpstead, Aleven, Koedinger · Advances in Cognitive Systems, 2016