FormalAnalyticGeo: A Neural-Symbolic Based Framework for Multimodal Analytic Geometry Problem Generation

Multimodal Large Language Models have improved math reasoning, but analytic geometry remains difficult due to a lack of annotated data. Current methods fail to produce the precise geometric constraints required for valid problems, often resulting in inaccurate diagrams.

The new framework combines neural networks with symbolic logic to generate problems and diagrams automatically. This approach ensures geometric elements, such as conic curves, are mathematically accurate and adhere to specific constraints.

By creating a scalable pipeline for data generation, this work provides a new resource for training and evaluating models on complex visual math tasks. It bridges the gap between generative AI and formal mathematical rigor.