Reproduce “Learning to Compute Gröbner Bases” 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
Hiroshi Kera, Yuki Ishihara, Yuta Kambe, Tristan Vaccon, Kazuhiro Yokoyama
Researchers pioneered the application of Transformer models to the NP-hard problem of Gr¨obner basis computation for 0-dimensional radical ideals, utilizing novel algebraic methods for efficient dataset generation and a hybrid input embedding. The approach achieved over 86% accuracy on rational coefficients with up to five variables and demonstrated the ability to solve some complex instances significantly faster than traditional algorithms.
Key insights
- Transformers can learn to perform Gr¨obner basis computation for specific classes of ideals (0-dimensional radical ideals in shape position), demonstrating learnability for an NP-hard problem.
- Efficient backward dataset generation is crucial for applying machine learning to computationally intensive mathematical problems where forward data generation is intractable.
- The models exhibited higher accuracy on infinite fields (Q, R) than on finite fields (Fp) and were more adept at predicting the polynomial structure (support) than precise coefficient values.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.