Skip to main content Skip to navigation

Amortised Design Optimization for Item Response Theory

Project Overview

The document explores the integration of generative AI into education, highlighting the innovative Amortised Design Optimization for Item Response Theory (ADOIRT) method, which leverages Deep Reinforcement Learning (DRL) to enhance the assessment of student abilities and optimize learning experiences. By utilizing Item Response Theory (IRT), ADOIRT facilitates near-real-time interactions with students, effectively selecting the most informative test items based on their previous performance. This approach not only streamlines the assessment process but also improves the calibration of educational tools, thereby personalizing learning experiences. The findings suggest that such AI-driven methods can significantly enhance the efficiency and effectiveness of educational assessments, ultimately leading to better learning outcomes and a more tailored educational environment for students.

Key Applications

Amortised Design Optimization for Item Response Theory (ADOIRT)

Context: Real-time assessment of student abilities through testing in educational settings.

Implementation: Incorporates amortised experimental design into IRT using a Deep Reinforcement Learning agent trained on synthetic data to optimize test item selection.

Outcomes: Improved accuracy in estimating student abilities with lower error rates compared to non-adaptive and random designs.

Challenges: High computational costs in initial training phase and the need for quality synthetic data for training.

Implementation Barriers

Technical

High computational cost associated with Optimal Experimental Design methods and the need for extensive pre-computation.

Proposed Solutions: Incorporating amortised methods and training on synthetic datasets to reduce computation during real-time interactions.

Data-related

Dependence on the quality of synthetic data for training the DRL agent, which may not fully represent real-world scenarios.

Proposed Solutions: Exploring real-world testing and refining the model based on actual student interactions.

Project Team

Antti Keurulainen

Researcher

Isak Westerlund

Researcher

Oskar Keurulainen

Researcher

Andrew Howes

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Antti Keurulainen, Isak Westerlund, Oskar Keurulainen, Andrew Howes

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies