meta-learning-evolution

What problem does it solve?

Design meta-learning systems that automatically evolve code, configurations, and prompts using genetic algorithms and large language models, enabling faster discovery and automation of outer-loop optimization workflows.

Core Features & Use Cases

• Level-1 GA-guided evolution for outer-loop optimization of loss functions, rewards, and hyperparameters.
• Level-2 LLM-driven prompt and primitive discovery to supplement GA search and inject new capabilities when stagnation occurs.
• A two-level evolution workflow with explicit evaluation metrics, reproducible experiment templates, and guidance for when to apply GA vs LLM.

Quick Start

Provide a two-level evolution plan that uses a genetic algorithm for outer-loop optimization and an LLM to invent new primitives when stagnation occurs, with explicit prompts and evaluation metrics.