Training Numerical Intelligence via Auto-Diagnosis and Skill Discovery

License: See LICENSE file · Updated 2026-10-06
My rating

This paper introduces Auto-Diagnosis and Skill Discovery (ADSD), a framework for scientific-computing numerical solvers that uses execution feedback to diagnose shortcomings and discover reusable improvement skills. The method is designed to help AI agents move beyond generating scientific code toward systematically improving the numerical algorithms that underlie it.

Tags: