Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Exploring Knowledge Transfer in Evolutionary Many-task Optimization: A Complex Network Perspective

Yudong Yang, Kai Wu, Xiangyi Teng, Handing Wang, He Yu, Jing Liu

TL;DR

This work reframes evolutionary many-task optimization as a networked problem, proposing a Knowledge Transfer Relationship Network (KTRN) to model how tasks exchange knowledge. By extracting and analyzing directed transfer networks from multiple EMaTO algorithms, it shows that KTRNs exhibit community structure, varying density, and hub-like heterogeneity, which jointly influence transfer efficacy and potential negative transfer. The study confirms that multi-population EMaTO methods generally outperform multi-factorial ones under the tested benchmarks and demonstrates how hyperparameters (e.g., cluster count and transfer breadth) modulate network density and transfer outcomes. It advocates leveraging complex-network techniques, like community detection, to refine KT strategies and reduce negative transfer, with practical implications for designing more scalable and efficient EMaTO systems.

Abstract

The field of evolutionary many-task optimization (EMaTO) is increasingly recognized for its ability to streamline the resolution of optimization challenges with repetitive characteristics, thereby conserving computational resources. This paper tackles the challenge of crafting efficient knowledge transfer mechanisms within EMaTO, a task complicated by the computational demands of individual task evaluations. We introduce a novel framework that employs a complex network to comprehensively analyze the dynamics of knowledge transfer between tasks within EMaTO. By extracting and scrutinizing the knowledge transfer network from existing EMaTO algorithms, we evaluate the influence of network modifications on overall algorithmic efficacy. Our findings indicate that these networks are diverse, displaying community-structured directed graph characteristics, with their network density adapting to different task sets. This research underscores the viability of integrating complex network concepts into EMaTO to refine knowledge transfer processes, paving the way for future advancements in the domain.

Exploring Knowledge Transfer in Evolutionary Many-task Optimization: A Complex Network Perspective

TL;DR

This work reframes evolutionary many-task optimization as a networked problem, proposing a Knowledge Transfer Relationship Network (KTRN) to model how tasks exchange knowledge. By extracting and analyzing directed transfer networks from multiple EMaTO algorithms, it shows that KTRNs exhibit community structure, varying density, and hub-like heterogeneity, which jointly influence transfer efficacy and potential negative transfer. The study confirms that multi-population EMaTO methods generally outperform multi-factorial ones under the tested benchmarks and demonstrates how hyperparameters (e.g., cluster count and transfer breadth) modulate network density and transfer outcomes. It advocates leveraging complex-network techniques, like community detection, to refine KT strategies and reduce negative transfer, with practical implications for designing more scalable and efficient EMaTO systems.

Abstract

The field of evolutionary many-task optimization (EMaTO) is increasingly recognized for its ability to streamline the resolution of optimization challenges with repetitive characteristics, thereby conserving computational resources. This paper tackles the challenge of crafting efficient knowledge transfer mechanisms within EMaTO, a task complicated by the computational demands of individual task evaluations. We introduce a novel framework that employs a complex network to comprehensively analyze the dynamics of knowledge transfer between tasks within EMaTO. By extracting and scrutinizing the knowledge transfer network from existing EMaTO algorithms, we evaluate the influence of network modifications on overall algorithmic efficacy. Our findings indicate that these networks are diverse, displaying community-structured directed graph characteristics, with their network density adapting to different task sets. This research underscores the viability of integrating complex network concepts into EMaTO to refine knowledge transfer processes, paving the way for future advancements in the domain.
Paper Structure (26 sections, 8 equations, 3 figures, 6 tables)

This paper contains 26 sections, 8 equations, 3 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Formulation of Multi-Task Optimization Problems
  4. A General Framework of EMaTO
  5. Networks
  6. Advancements in EMaTO Algorithms
  7. Algorithmic Classification and Comparison
  8. Multi-factorial Approach
  9. Multi-Population Approach
  10. Algorithm Overview
  11. Experiments
  12. Benchmark
  13. Optimization Performance Comparison
  14. Evaluating KTRN in EMaTO
  15. Capturing KTRN in Algorithm Execution
  16. ...and 11 more sections

Figures (3)

  • Figure 1: General process of knowledge transfer guided by KTRN in EMaTO algorithm: an illustrative diagram. This schematic represents the initialization of separate populations for each task within multi-swarm algorithms. It evaluates inter-population correlations to generate a directed network indicating knowledge transfer relationships, with each edge signifying a knowledge transfer event. The module on the right demonstrates the universal framework for knowledge transfer: extracting beneficial knowledge from an assisted population using algorithm-specific mechanisms, transforming it for the target population, and thereby injecting it to facilitate the transfer of general experience during the evolutionary process.
  • Figure 2: Convergence performance comparison of six renowned EMaTO algorithms versus single-task DE algorithm on the WCCI-20 MaTOP benchmark
  • Figure 3: Convergence performance comparison of the EMaTO-MKT algorithm on select tasks from the WCCI20 MaTOP benchmark under various parameter combinations.