Reproduce “Data Augmentation for Mathematical Objects” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tereso del Rio, Matthew England
Researchers at Coventry University enhanced the performance of machine learning models for Cylindrical Algebraic Decomposition (CAD) variable ordering, demonstrating that data augmentation through variable renaming and strategic dataset balancing leads to an average 63% increase in model accuracy and improved generalization across different data distributions.
Key insights
- Exploiting inherent symmetries in mathematical objects, such as arbitrary variable naming, enables cost-effective data augmentation without needing additional expensive computations for new labels.
- Correcting class imbalance in training datasets is crucial for improving ML model generalization, even when the overall dataset size remains constant.
- The most substantial performance gains and model robustness for symbolic computation tasks are achieved by combining both dataset balancing and a significant increase in data size through intelligent augmentation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.