Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evolutionary Computation for the Design and Enrichment of General-Purpose Artificial Intelligence Systems: Survey and Prospects

Javier Poyatos, Javier Del Ser, Salvador Garcia, Hisao Ishibuchi, Daniel Molina, Isaac Triguero, Bing Xue, Xin Yao, Francisco Herrera

TL;DR

The paper surveys how Evolutionary Computation can drive the design and enrichment of General-Purpose AI Systems (GPAIS), emphasizing open-world adaptability and multitask capabilities. It introduces a taxonomy of EC-powered AI, maps GPAIS properties to ML and EC research areas, and reviews milestones from closed-world neural architecture search to open-ended, diversity-driven approaches. Notable contributions include identifying EC-enabled design categories (hyper-parameter optimization, automated algorithm selection/design, algorithm construction) and enrichment strategies (diversity generation, data synthesis, learning-to-learn, active learning, collaboration). The discussion highlights challenges such as data scarcity, expensive evaluations, and balancing competing objectives, and outlines strategies like surrogate-assisted optimization, transfer learning, quality-diversity, and open-ended evolution to advance GPAIS toward self-constructing and self-adapting systems with practical impact.

Abstract

In Artificial Intelligence, there is an increasing demand for adaptive models capable of dealing with a diverse spectrum of learning tasks, surpassing the limitations of systems devised to cope with a single task. The recent emergence of General-Purpose Artificial Intelligence Systems (GPAIS) poses model configuration and adaptability challenges at far greater complexity scales than the optimal design of traditional Machine Learning models. Evolutionary Computation (EC) has been a useful tool for both the design and optimization of Machine Learning models, endowing them with the capability to configure and/or adapt themselves to the task under consideration. Therefore, their application to GPAIS is a natural choice. This paper aims to analyze the role of EC in the field of GPAIS, exploring the use of EC for their design or enrichment. We also match GPAIS properties to Machine Learning areas in which EC has had a notable contribution, highlighting recent milestones of EC for GPAIS. Furthermore, we discuss the challenges of harnessing the benefits of EC for GPAIS, presenting different strategies to both design and improve GPAIS with EC, covering tangential areas, identifying research niches, and outlining potential research directions for EC and GPAIS.

Evolutionary Computation for the Design and Enrichment of General-Purpose Artificial Intelligence Systems: Survey and Prospects

TL;DR

The paper surveys how Evolutionary Computation can drive the design and enrichment of General-Purpose AI Systems (GPAIS), emphasizing open-world adaptability and multitask capabilities. It introduces a taxonomy of EC-powered AI, maps GPAIS properties to ML and EC research areas, and reviews milestones from closed-world neural architecture search to open-ended, diversity-driven approaches. Notable contributions include identifying EC-enabled design categories (hyper-parameter optimization, automated algorithm selection/design, algorithm construction) and enrichment strategies (diversity generation, data synthesis, learning-to-learn, active learning, collaboration). The discussion highlights challenges such as data scarcity, expensive evaluations, and balancing competing objectives, and outlines strategies like surrogate-assisted optimization, transfer learning, quality-diversity, and open-ended evolution to advance GPAIS toward self-constructing and self-adapting systems with practical impact.

Abstract

In Artificial Intelligence, there is an increasing demand for adaptive models capable of dealing with a diverse spectrum of learning tasks, surpassing the limitations of systems devised to cope with a single task. The recent emergence of General-Purpose Artificial Intelligence Systems (GPAIS) poses model configuration and adaptability challenges at far greater complexity scales than the optimal design of traditional Machine Learning models. Evolutionary Computation (EC) has been a useful tool for both the design and optimization of Machine Learning models, endowing them with the capability to configure and/or adapt themselves to the task under consideration. Therefore, their application to GPAIS is a natural choice. This paper aims to analyze the role of EC in the field of GPAIS, exploring the use of EC for their design or enrichment. We also match GPAIS properties to Machine Learning areas in which EC has had a notable contribution, highlighting recent milestones of EC for GPAIS. Furthermore, we discuss the challenges of harnessing the benefits of EC for GPAIS, presenting different strategies to both design and improve GPAIS with EC, covering tangential areas, identifying research niches, and outlining potential research directions for EC and GPAIS.
Paper Structure (17 sections, 2 figures, 2 tables)

This paper contains 17 sections, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Evolutionary Computation and GPAIS
  3. Evolutionary Computation in Machine Learning
  4. GPAIS: Definitions and Properties
  5. AI-powered AI for GPAIS
  6. Matching the potential of EC with GPAIS: A Taxonomy for EC-powered AI and notable milestones
  7. EC-powered AI for designing GPAIS
  8. EC-powered AI for enriching GPAIS
  9. Connection among GPAIS, Machine Learning, and Optimization Research areas
  10. Notable milestones and achievements in EC-GPAIS
  11. Challenges of harnessing the EC-GPAIS benefits and strategies to address advances
  12. Challenges of harnessing the benefits of EC-GPAIS
  13. EC-GPAIS by exploiting and adapting existing knowledge
  14. EC-GPAIS by continuously creating new knowledge
  15. EC-GPAIS by constructing new models from scratch
  16. ...and 2 more sections

Figures (2)

  • Figure 1: Closed-world vs. Open-world GPAIS (adapted from Triguero2024).
  • Figure 3: Summary of the different strategies to leverage EC in the design and enrichment of GPAIS, together with the EC research areas related to each strategy. Problems undergone by the adoption of EC are common to all strategies, and the research areas studying how to overcome them are also included.