Landscape-aware Automated Algorithm Design: An Efficient Framework for Real-world Optimization

TL;DR

Real-world optimization often incurs prohibitive evaluation costs, hindering automated algorithm design. The paper proposes a landscape-aware framework that decouples discovery from expensive evaluations by generating proxy functions with Genetic Programming that mimic the target landscape through $W_1$-based alignment on Exploratory Landscape Analysis features, enabling an LLM-driven designer to search on proxies. Empirical results across meta-surface, Bragg mirrors, ellipsometry, and photovoltaic design show proxy-driven discovery achieves high-performance algorithms with substantially fewer real evaluations, often outperforming baseline methods. This approach demonstrates that landscape-aligned proxies can enable practical, resource-efficient LLM-driven algorithm design for computationally intensive real-world optimization problems.

Abstract

The advent of Large Language Models (LLMs) has opened new frontiers in automated algorithm design, giving rise to numerous powerful methods. However, these approaches retain critical limitations: they require extensive evaluation of the target problem to guide the search process, making them impractical for real-world optimization tasks, where each evaluation consumes substantial computational resources. This research proposes an innovative and efficient framework that decouples algorithm discovery from high-cost evaluation. Our core innovation lies in combining a Genetic Programming (GP) function generator with an LLM-driven evolutionary algorithm designer. The evolutionary direction of the GP-based function generator is guided by the similarity between the landscape characteristics of generated proxy functions and those of real-world problems, ensuring that algorithms discovered via proxy functions exhibit comparable performance on real-world problems. Our method enables deep exploration of the algorithmic space before final validation while avoiding costly real-world evaluations. We validated the framework's efficacy across multiple real-world problems, demonstrating its ability to discover high-performance algorithms while substantially reducing expensive evaluations. This approach shows a path to apply LLM-based automated algorithm design to computationally intensive real-world optimization challenges.

Figures (4)

• Figure 1: An overview of the proposed landscape-aware AAD framework. It first generates low-cost proxy functions using GP to match the landscape of the real problem, then uses LLMs to discover algorithms on these proxies, and finally validates only top candidates on the real world problem.
• Figure 2: The distribution of AOCC values across 10 runs of different algorithms on real-world problems, where higher values are better. The $x$-axis represents different algorithms, and the $y$-axis represents the AOCC value. The algorithms in Figure \ref{['fig:AOCC_meta_surface_surrogate']} run on the surrogate model of meta-surface, while the algorithms in Figure \ref{['fig:AOCC_meta_surface_solver']} run on the real solver of meta-surface.
• Figure 3: Benchmark results of the generated and baseline algorithms. The $x$-axis represents the evaluations of the problem and the $y$-axis represents the normalized fitness value, which are all smaller, the better. Each curve is averaged over 10 runs. For most real-world problems, algorithms generated with proxy functions demonstrate performance similar to algorithms derived from real-world problems while simultaneously exhibiting marked superiority over algorithms based on artificially designed problem discovery.
• Figure 4: Best solutions found by proxy-driven algorithms.