Evolutionary Generative Optimization: Towards Fully Data-Driven Evolutionary Optimization via Generative Learning

TL;DR

EvoGO introduces a fully data-driven evolutionary optimization framework that autonomously learns the entire search process using generative models, significantly improving convergence speed and performance across diverse tasks.

Contribution

This work presents EvoGO, a novel framework that replaces handcrafted operators with generative models, enabling fully data-driven and autonomous evolutionary optimization.

Findings

EvoGO converges within 10 generations on various benchmarks.

EvoGO outperforms traditional EAs, Bayesian optimization, and RL methods.

The approach is effective on numerical, control, and robotic tasks.

Abstract

Recent advances in data-driven evolutionary algorithms (EAs) have demonstrated the potential of leveraging historical data to improve optimization accuracy and adaptability. Despite these advancements, existing methods remain reliant on handcrafted process-level operators. In contrast, Evolutionary Generative Optimization (EvoGO) is a fully data-driven framework designed from the objective level, enabling autonomous learning of the entire search process. EvoGO streamlines the evolutionary optimization process into three stages: data preparation, model training, and population generation. The data preparation stage constructs a pairwise dataset to enrich training diversity without incurring additional evaluation costs. During model training, a tailored generative model learns to transform inferior solutions into superior ones. In the population generation stage, EvoGO replaces traditional reproduction operators with a scalable and parallelizable generative mechanism. Extensive experiments on numerical benchmarks, classical control problems, and high-dimensional robotic tasks demonstrate that EvoGO consistently converges within merely 10 generations and substantially outperforms a wide spectrum of optimization approaches, including traditional EAs, Bayesian optimization, and reinforcement learning based methods. Code is available at: https://github.com/EMI-Group/evogo