NeurGO: Learning to Generate Elite Candidates for Meta-Black-Box Expensive Optimization

2026-07-25

AI Summary

This work addresses the challenges of expensive black-box optimization, where high evaluation costs, limited budgets, and the tendency of conventional methods to waste resources on poor-quality solutions hinder performance. To overcome these issues, the authors propose a generative meta black-box optimization framework that leverages an attention-based encoder to capture the search dynamics of historical populations and conditionally generates high-quality candidate solutions. This approach bypasses the need for global surrogate models, thereby avoiding associated approximation biases and risks of premature convergence to local optima. A joint quality-diversity loss function is employed to simultaneously ensure convergence efficiency and maintain population diversity. Empirical evaluations on the CEC 2008 and COCO BBOB benchmark suites demonstrate that the proposed method significantly outperforms existing approaches under identical evaluation budgets, achieving faster convergence and superior final solution quality.

Abstract

Expensive black-box optimization is ubiquitous in science and engineering, where function evaluations are costly and the evaluation budget is limited. Traditional evolutionary algorithms and Meta-BlackBox Optimization (MetaBBO) approaches typically consume most evaluations on candidate selection, often wasting precious budget on inferior solutions. Although surrogate-assisted evolution and Bayesian optimization aim to reduce evaluations through surrogate models, constructing an accurate global model from limited data remains challenging, and model bias can easily trap the search in local optima. To overcome these limitations, we propose NeurGO, a generative MetaBBO framework that directly synthesizes elite candidates from historical population states. Specifically, we employ an attention-based encoder to capture the population-level search trend and condition a decoder on this representation to generate high-quality candidates, avoiding the expensive evaluation of large offspring pools. We then design a quality-diversity loss to maintain solution quality and population diversity throughout the search. Through extensive benchmarking on CEC 2008 and the COCO BBOB test suites, our method achieves better optimization performance under the same evaluation budget and exhibits faster convergence.

Problem

• expensive black-box optimization
• Meta-Black-Box Optimization
• surrogate-assisted evolution
• Bayesian optimization
• evaluation budget

Innovation

• generative MetaBBO
• attention-based encoder
• elite candidate generation
• quality-diversity loss
• surrogate-free optimization