Skip to main content Skip to navigation

MNIST-Fraction: Enhancing Math Education with AI-Driven Fraction Detection and Analysis

Project Overview

The document introduces MNIST-Fraction, an innovative dataset aimed at improving mathematics education through the application of generative AI, specifically via deep learning techniques like Convolutional Neural Networks (CNNs) for fraction detection and analysis. This dataset addresses the complexities associated with recognizing various handwritten representations of fractions, enabling its use in automated grading systems, document analysis, and broader educational technology applications. The findings highlight notable advancements in both the accuracy and efficiency of recognizing fractions in educational settings, suggesting that such AI-driven tools can significantly enhance the learning experience and administrative processes in mathematics education. Overall, the integration of generative AI in educational contexts, particularly through the use of MNIST-Fraction, showcases its potential to transform traditional approaches to teaching and assessing mathematical concepts.

Key Applications

MNIST-Fraction dataset for fraction detection

Context: Educational technology, aimed at educators and students learning fractions

Implementation: Developed a dataset of synthetic fraction images from MNIST; trained CNNs for recognition tasks using this dataset.

Outcomes: Improved accuracy and efficiency in detecting and analyzing handwritten fractions; supports automated grading and personalized feedback.

Challenges: Difficulties in detecting variations in handwriting styles and formats; reliance on high-quality datasets.

Implementation Barriers

Data Quality

Scarcity of quality datasets for training AI models in recognizing handwritten mathematical expressions.

Proposed Solutions: Creation of the MNIST-Fraction dataset tailored for fraction recognition, ensuring comprehensive coverage of various styles.

Technical Complexity

Challenges in processing handwritten information due to diverse fonts, styles, and formats.

Proposed Solutions: Utilization of deep learning techniques (CNNs) to enhance detection capabilities and accuracy.

Project Team

Pegah Ahadian

Researcher

Yunhe Feng

Researcher

Karl Kosko

Researcher

Richard Ferdig

Researcher

Qiang Guan

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Pegah Ahadian, Yunhe Feng, Karl Kosko, Richard Ferdig, Qiang Guan

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