Astra AI
Joined 2023
Partners
University of Memphis
Carnegie Learning
What are we building?
This project is developing ASTRA (AI-based STRategy Discovery) with the goal of creating Large Strategy Models (LSMs) that encode Math problem solving strategies. Analogous to Large Language Models (LLMs) that understand language structure, our goal is to learn structures underlying problem solving in Math. Using this, we aim to build a general research platform that can be utilized to solve several challenging tasks in education such as inferring mastery, designing personalized interventions, etc.
What are we learning?
How to use state-of-the-art AI methods to discover patterns in strategy use from large-scale interaction data collected from a diverse set of schools and students.
How to use these patterns to understand if and how students develop problem-solving strategies that are context-specific and efficient.
Products
Novel, open access AI models that encode strategies using large-scale datasets collected from student-ITS interactions.
Visualization tools to analyze strategies at scale
AIMS Collaboratory | Inventory of Public Goods
Public goods shared by the Astra AI project team:
For: Researchers, Developers
Analyzing Strategies in MATHia with BERT
Uses large-scale MATHia data and BERT models to analyze how students choose and adapt math strategies across problem contexts.
Academic Article
For: Researchers, Developers
Analyzing Math Learning Strategies at Scale using AI Representations
Presents the ASTRA approach, using AI representation learning to analyze math strategies in large-scale student interaction data from MATHia.
Presentation/Poster
For: Researchers, Developers
“Can A Language Model Represent Math Strategies?”: Learning Math Strategies from Big Data using BERT
Shows how BERT can learn representations of student math strategies from MATHia action sequences and predict strategy use on new problems.
Research Paper
