# Keplearn reports equation recovery gains with limited curve-fitting performance

_Published Thursday, October 29, 2026 at 5:19 PM EDT · Science · Latest · Tier 2 — Notable_

![Symbolic regression in seconds. Contribute to owls-on-wires/keplearn development by creating an account on GitHub. — Primary](https://opengraph.githubassets.com/ac933e9108f082837b9b294a0971004683f1e63c449d8328b8217689e18faa81/owls-on-wires/keplearn)

Keplearn, an MIT-licensed tool that searches measurement tables for mathematical equations, reports recovering exact symbolic solutions on 64.7% of 116 noise-free Feynman benchmark datasets. Its project documentation compares that result with AIFeynman's published 55.8% rate.

The Rust program uses recursive mathematical transformations and checks candidate equations against held-out data. Binaries are available for macOS, Windows and Linux. The reported benchmark used one trial per dataset and a 60-second fitting budget. Performance deteriorated with noise, and median test R2 was about -0.003 across 138 datasets without known closed-form equations.

## Sources

- [github.com](https://github.com/owls-on-wires/keplearn)

---
Canonical: https://techandbusiness.org/newswire/7MdIr8iovVtI6V-MFS-tkY
Published: 2026-10-29T21:19:09.582Z
Story chronology: 2026-10-29T14:22:53.000Z
Retrieved: 2026-10-29T23:37:58.653Z
Publisher: Tech & Business (techandbusiness.org)