Research map / Learning theory and algorithms / Combinatorial and search optimization
AutoNumerics-Zero: Automated Discovery of State-of-the-Art Mathematical Functions
Esteban Real, Mirko Rossini, Connal de Souza, Manav Garg, Moritz Firsching, Quoc V Le, Yao Chen, Akhil Verghese, Ekin Dogus Cubuk, David H. Park
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace NSGA-II selection with random search (ablation) | AutoNumerics-Zero · 2^x search | accuracy vs operations | Hurt |
| add CMA-ES coefficient optimization (ablation) | AutoNumerics-Zero · 2^x search | accuracy vs operations | Improved |
| use guided mutations of the compute graph instead of random graphs (ablation) | AutoNumerics-Zero · 2^x search | accuracy vs operations | Improved |
| search for programs from scratch instead of Taylor/Pade/Chebyshev/minimax forms | exponential 2^x · 2^x on R via range reduction | accuracy for same number of operations: 14 significant figures with 10 operations | Improved |
| evolved programs versus Pade approximants at higher accuracies | AutoNumerics-Zero · erf on (0,2] | operation count at given accuracy | Mixed |
| evolve against hardware speed with float32 and ULP error objective | AutoNumerics-Zero · 2^x hardware-aware (Skylake) | throughput speed: more than 3 times faster | Improved |
| evolved Airy program versus best 20-operation baseline | AutoNumerics-Zero · Airy Ai(-7x) | max error: two orders of magnitude lower error | Improved |
| hand-designed Chebyshev baseline reparameterized for Bessel | Chebyshev baseline · Bessel I_1/2(x) | accuracy: narrows gap to 7.8 vs 8.1 evolved | Improved |
Closest papers by meaning
- Unsupervised Discovery of Formulas for Mathematical Constants (NeurIPS 2024 · similarity 0.87)
- LLM-SR: Scientific Equation Discovery via Programming with Large Language Models (ICLR 2025 · similarity 0.85)
- FunctionEvolve: Structure-Guided Symbolic Regression with LLMs (NeurIPS 2026 · similarity 0.84)
- Improving Monte Carlo Tree Search for Symbolic Regression (NeurIPS 2025 · similarity 0.84)
- Symbolic regression via MDLformer-guided search: from minimizing prediction error to minimizing description length (ICLR 2025 · similarity 0.84)
- FIND: A Framework for Discovering Formulas in Data (AAAI 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
