Skip to content

Adaptive reinforcement-learning tutoring system modeling cross-cultural competence through Dewey-inspired state–action–reward design. Developed for CSCE 642 (TAMU).

License

Notifications You must be signed in to change notification settings

aggiewissenschaftler/PedagoReLearn

Repository files navigation

PedagoReLearn

© 2025 Thomas F. Hallmark
Licensed under the MIT License


Overview

PedagoReLearn is an AI-driven reinforcement learning (RL) framework for adaptive cross-cultural tutoring.
It models cultural competence training as a state–action–reward process where an agent learns when to teach, review, or quiz learners across domains such as etiquette, privacy, work, and travel.

Grounded in John Dewey’s progressive education theory, the system demonstrates how pedagogical strategies can emerge autonomously through experience, advancing long-term mastery and retention.


Project Summary

PedagoReLearn formulates tutoring as a Markov Decision Process (MDP):

Component Description
States Learner mastery levels and recency of review for each cultural rule
Actions teach, quiz, or review
Rewards Reflect learning success, retention, and teaching efficiency

Each YAML file under /rules/ defines a cultural knowledge domain (e.g., workplace etiquette, travel behavior, hygiene norms).
The Gymnasium environment (env.env.py) interprets these as interactive learning topics.


Repository Structure

PedagoReLearn/
│	# -------------------- Documents --------------------
├── _archive/						# Archive of the whole semester
├── rules/							# YAML cultural knowledge base
│
│	#------------------    Codes    --------------------
├── agents/tutor.py                 # RL Tutor agents implementations. includes:
│			            			# 	SARSA(0) on-policy learner
│   					          	# 	Random baseline policy
│ 									#	Fixed baseline policy
├── env/env.py                      # Core Gymnasium environment
├── students/student.py             # Simulated Student
├── utils/                          # Utilities
│   ├── data_util.py				# 	Full result data structure & logger
│   └── plotter.py					# 	Full result figure generator
│
├── trainer.py                      # RL tutor trainer
├── main_train.py                   # Main script for train & result
│
│	#------------------   Results   --------------------
├── fig_results/                    # Figures of results
├── ts_results/                    	# Full results for each step in episode
│									# to generate figures
│
├── LICENSE
├── README.md
└── requirements.txt

Quick Start

Installation

# 0. Clone repository

# 1. Create and activate virtual environment
python -m venv .venv
source .venv/bin/activate        # macOS/Linux
# .venv\Scripts\activate          # Windows

# 2. Install dependencies
pip install -r requirements.txt

View Student model

# compare student's forgetting curve to Ebbinghaus Forgetting Curve
python -m students.student forgetting_curve

Train & Get result

# Train PedagoReLearn SARSA Tutor for n_seeds
# and compare with random and fixed policy tutor. 
python -m main_train --n_rules 3 --n_episodes 2000 --n_seeds 50 --ma_window 0.01 --track_full_result

Resulting figures can be found in fig_results/

Tech Stack

  • Python 3.10+
  • Gymnasium ≥ 0.29
  • NumPy ≥ 1.23
  • Matplotlib ≥ 3.7
  • PyYAML for rule parsing and validation

Future Directions

  • Full aggregation ablation across four schemes
  • Sensitivity analysis for α, γ, and ε-decay parameters
  • Integration of heuristic spaced-repetition baseline
  • Policy interpretability visualization (heatmaps, frequency plots)
  • Expansion to additional cultural domains and learner models

Acknowledgments

Developed for CSCE 642: Reinforcement Learning (Fall 2025) Texas A&M University

Conceptual design and implementation by Thomas F. Hallmark and Jun Kwon.

AUTHOR BIOGRAPHIES

Hallmark, T. F. (2025). | thomas.hallmark@tamu.edu

Thomas F. Hallmark is a doctoral student in Curriculum and Instruction with a cognate in Engineering Education in the Department of Teaching, Learning, and Culture at Texas A&M University. He holds degrees in Legal Studies and Business Administration (MBA) and brings more than 30 years of experience in the nuclear and utilities industries. His research focuses on the integration of artificial intelligence and reinforcement learning in engineering and STEM education, emphasizing adaptive tutoring systems, veteran transitions, and cross-cultural learning. Hallmark’s work combines pedagogical theory with computational modeling to design human-centered AI learning environments.

Kwon, J. (2025).

Jun Kwon is a graduate student in Computer Science and Engineering at Texas A&M University, specializing in machine learning and artificial intelligence applications for education and human-computer interaction. His research interests include reinforcement learning algorithms, neural network optimization, and adaptive feedback mechanisms in educational software. Kwon contributes to the computational architecture and algorithmic implementation of PedagoReLearn, focusing on model design, environment development, and performance evaluation across multiple RL frameworks.

Joint Contribution Hallmark and Kwon collaboratively developed the conceptual framework and technical implementation of PedagoReLearn, merging educational theory and AI engineering to advance research in adaptive tutoring systems and cultural-learning reinforcement models.

GitHub Description

Adaptive RL tutoring system modeling cultural learning through Dewey-inspired state, action, and reward design.

AI Use Disclaimer

Artificial intelligence (AI) tools—including ChatGPT—were used only for grammar, formatting, and document organization. All intellectual content (code, methodology, analysis) is the original work of the authors and complies with Texas A&M University academic integrity standards.

About

Adaptive reinforcement-learning tutoring system modeling cross-cultural competence through Dewey-inspired state–action–reward design. Developed for CSCE 642 (TAMU).

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •