Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Performance to Understanding: A Vision for Explainable Automated Algorithm Design

Niki van Stein, Anna V. Kononova, Thomas Bäck

TL;DR

Automated algorithm design with large language models offers powerful exploration but remains largely opaque. The paper advocates an explainable automated algorithm design (AAD) framework built on three pillars: LLM-driven discovery of algorithmic variants, explainable benchmarking that attributes performance to components and hyperparameters, and problem-class descriptors that link algorithm behavior to landscape structure. It envisions a closed knowledge loop—Discover, Explain, Describe, Generalise—where discoveries are rigorously evaluated, explanations guide generalisation, and descriptors enable class-specific rules, ultimately producing reusable scientific insights. By embedding explanation and problem structure into the discovery process, the approach aims to shift from blind search to interpretable, class-aware design with practical impact across real-world optimisation tasks.

Abstract

Automated algorithm design is entering a new phase: Large Language Models can now generate full optimisation (meta)heuristics, explore vast design spaces and adapt through iterative feedback. Yet this rapid progress is largely performance-driven and opaque. Current LLM-based approaches rarely reveal why a generated algorithm works, which components matter or how design choices relate to underlying problem structures. This paper argues that the next breakthrough will come not from more automation, but from coupling automation with understanding from systematic benchmarking. We outline a vision for explainable automated algorithm design, built on three pillars: (i) LLM-driven discovery of algorithmic variants, (ii) explainable benchmarking that attributes performance to components and hyperparameters and (iii) problem-class descriptors that connect algorithm behaviour to landscape structure. Together, these elements form a closed knowledge loop in which discovery, explanation and generalisation reinforce each other. We argue that this integration will shift the field from blind search to interpretable, class-specific algorithm design, accelerating progress while producing reusable scientific insight into when and why optimisation strategies succeed.

From Performance to Understanding: A Vision for Explainable Automated Algorithm Design

TL;DR

Automated algorithm design with large language models offers powerful exploration but remains largely opaque. The paper advocates an explainable automated algorithm design (AAD) framework built on three pillars: LLM-driven discovery of algorithmic variants, explainable benchmarking that attributes performance to components and hyperparameters, and problem-class descriptors that link algorithm behavior to landscape structure. It envisions a closed knowledge loop—Discover, Explain, Describe, Generalise—where discoveries are rigorously evaluated, explanations guide generalisation, and descriptors enable class-specific rules, ultimately producing reusable scientific insights. By embedding explanation and problem structure into the discovery process, the approach aims to shift from blind search to interpretable, class-aware design with practical impact across real-world optimisation tasks.

Abstract

Automated algorithm design is entering a new phase: Large Language Models can now generate full optimisation (meta)heuristics, explore vast design spaces and adapt through iterative feedback. Yet this rapid progress is largely performance-driven and opaque. Current LLM-based approaches rarely reveal why a generated algorithm works, which components matter or how design choices relate to underlying problem structures. This paper argues that the next breakthrough will come not from more automation, but from coupling automation with understanding from systematic benchmarking. We outline a vision for explainable automated algorithm design, built on three pillars: (i) LLM-driven discovery of algorithmic variants, (ii) explainable benchmarking that attributes performance to components and hyperparameters and (iii) problem-class descriptors that connect algorithm behaviour to landscape structure. Together, these elements form a closed knowledge loop in which discovery, explanation and generalisation reinforce each other. We argue that this integration will shift the field from blind search to interpretable, class-specific algorithm design, accelerating progress while producing reusable scientific insight into when and why optimisation strategies succeed.

Paper Structure

This paper contains 16 sections, 1 figure.

Table of Contents

  1. Introduction
  2. Past and Current Trends in EC
  3. A Brief History of Algorithm Design in EC
  4. Evolution of Benchmarking Practices
  5. Demonstration of applicability of the algorithm
  6. Extensive convergence plots
  7. Performance per function group
  8. Domain-specfic benchmarking
  9. Modern benchmarking
  10. Explainable Benchmarking
  11. Automated Algorithm Selection, Configuration and Design
  12. Motivation
  13. The Case for Class-Specific Discovery
  14. Integrating Discovery and Explanation: A Vision
  15. Implications and Research Agenda
  16. ...and 1 more sections

Figures (1)

  • Figure 1: General timeline of algorithm development in Evolutionary Computation -- a vision.