Advances in Mathematical Expression Recognition and Reasoning

The field of mathematical expression recognition and reasoning is witnessing significant advancements, driven by innovative applications of self-supervised learning, reinforcement learning, and large language models. Researchers are exploring novel approaches to improve the recognition of handwritten mathematical expressions, such as using progressive spatial masking strategies and context-aware voice-powered math workspaces. Additionally, there is a growing focus on developing compact and efficient language models optimized for specific academic exams, such as the Joint Entrance Examination (JEE). These models are being fine-tuned using curriculum learning and reinforcement learning with verifiable rewards to boost their performance. Noteworthy papers in this area include Mask & Match, which presents a self-supervised learning framework for recognizing handwritten mathematical expressions, and Aryabhata, which introduces a compact 7B parameter math reasoning model optimized for the JEE. We-Math 2.0 is also a notable contribution, as it integrates a structured mathematical knowledge system, model-centric data space modeling, and a reinforcement learning-based training paradigm to enhance the mathematical reasoning abilities of multimodal large language models.

Sources

Mask & Match: Learning to Recognize Handwritten Math with Self-Supervised Attention

Phoenix: A Novel Context-Aware Voice-Powered Math Equation Workspace and Editor

Aryabhata: An exam-focused language model for JEE Math

Compass-Thinker-7B Technical Report

Towards Scalable Training for Handwritten Mathematical Expression Recognition

We-Math 2.0: A Versatile MathBook System for Incentivizing Visual Mathematical Reasoning

Built with on top of