Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Novel Population Initialization Method via Adaptive Experience Transfer for General-Purpose Binary Evolutionary Optimization

Zhiyuan Wang, Shengcai Liu, Shaofeng Zhang, Ke Tang

TL;DR

The paper tackles the challenge of producing high-quality initial populations for binary evolutionary algorithms under tight evaluation budgets by proposing MPI, a general-purpose, cross-problem-class population initializer. MPI builds an offline repository of solving experiences, represents them with a VAE-based surrogate, and uses a gating network to select relevant experiences for a new instance, followed by target-specific transfer via decoder fine-tuning and extensive candidate generation. Experiments across six binary problem classes (three classic, three unseen) show MPI consistently improves performance over state-of-the-art PI methods, with strong transfer to higher dimensions and completely new problem classes. The work demonstrates a scalable, problem-agnostic strategy for transfer learning in discrete optimization, with practical implications for scenarios where function evaluations are expensive or time-constrained.

Abstract

Evolutionary Algorithms (EAs) are widely used general-purpose optimization methods due to their domain independence. However, under a limited number of function evaluations (#FEs), the performance of EAs is quite sensitive to the quality of the initial population. Obtaining a high-quality initial population without problem-specific knowledge remains a significant challenge. To address this, this work proposes a general-purpose population initialization method, named mixture-of-experience for population initialization (MPI), for binary optimization problems where decision variables take values of 0 or 1. MPI leverages solving experiences from previously solved problems to generate high-quality initial populations for new problems using only a small number of FEs. Its main novelty lies in a general-purpose approach for representing, selecting, and transferring solving experiences without requiring problem-specific knowledge. Extensive experiments are conducted across six binary optimization problem classes, comprising three classic classes and three complex classes from real-world applications. The experience repository is constructed solely based on instances from the three classic classes, while the performance evaluation is performed across all six classes. The results demonstrate that MPI effectively transfers solving experiences to unseen problem classes (i.e., the complex ones) and higher-dimensional problem instances, significantly outperforming existing general-purpose population initialization methods.

A Novel Population Initialization Method via Adaptive Experience Transfer for General-Purpose Binary Evolutionary Optimization

TL;DR

The paper tackles the challenge of producing high-quality initial populations for binary evolutionary algorithms under tight evaluation budgets by proposing MPI, a general-purpose, cross-problem-class population initializer. MPI builds an offline repository of solving experiences, represents them with a VAE-based surrogate, and uses a gating network to select relevant experiences for a new instance, followed by target-specific transfer via decoder fine-tuning and extensive candidate generation. Experiments across six binary problem classes (three classic, three unseen) show MPI consistently improves performance over state-of-the-art PI methods, with strong transfer to higher dimensions and completely new problem classes. The work demonstrates a scalable, problem-agnostic strategy for transfer learning in discrete optimization, with practical implications for scenarios where function evaluations are expensive or time-constrained.

Abstract

Evolutionary Algorithms (EAs) are widely used general-purpose optimization methods due to their domain independence. However, under a limited number of function evaluations (#FEs), the performance of EAs is quite sensitive to the quality of the initial population. Obtaining a high-quality initial population without problem-specific knowledge remains a significant challenge. To address this, this work proposes a general-purpose population initialization method, named mixture-of-experience for population initialization (MPI), for binary optimization problems where decision variables take values of 0 or 1. MPI leverages solving experiences from previously solved problems to generate high-quality initial populations for new problems using only a small number of FEs. Its main novelty lies in a general-purpose approach for representing, selecting, and transferring solving experiences without requiring problem-specific knowledge. Extensive experiments are conducted across six binary optimization problem classes, comprising three classic classes and three complex classes from real-world applications. The experience repository is constructed solely based on instances from the three classic classes, while the performance evaluation is performed across all six classes. The results demonstrate that MPI effectively transfers solving experiences to unseen problem classes (i.e., the complex ones) and higher-dimensional problem instances, significantly outperforming existing general-purpose population initialization methods.

Paper Structure

This paper contains 34 sections, 17 equations, 6 figures, 6 tables, 5 algorithms.

Table of Contents

  1. Introduction
  2. Population Initialization for EAs
  3. Notations and Problem Description
  4. Existing Population Initialization Methods
  5. Experience Transfer for Population Initialization
  6. Offline Experience Repository Preparation
  7. Step 1: Solving Experience Collection
  8. Step 2: Uniform Representation of Solving Experience
  9. Online Initial Population Generation
  10. Step 1: Solving Experience Selection
  11. Step 2: Solving Experience Transfer
  12. Step 3: Initial Population Generation
  13. Gating Network Training
  14. Computational Studies
  15. Problem Instances and Constraint Handling
  16. ...and 19 more sections

Figures (6)

  • Figure 1: An illustration for the workflow of the proposed MPI method.
  • Figure 2: The overall pipeline of MPI, detailing the steps within the offline and online phases.
  • Figure 3: An illustration for the structure of the surrogate model that represents the solving experience for a problem instance.
  • Figure 4: Performance comparison of MPI VS. the Baselines across #FE budgets. X-axis: #FEs; Y-axis: performance metrics. #Avg. $\uparrow$: the number of instances where the MPI's average objective value of the best solution (from 30 runs) exceeds the Baseline's. Goal Diff: Win number minus Loss number from statistical testing (W-D-L results) on best solutions over 30 runs. Each subplot includes a red horizontal line dividing the area into Better (MPI outperforms the Baseline) and Worse (the Baseline outperforms MPI) areas. The subfigures present: (a) #Avg. $\uparrow$ metric on Elite-GA; (b) Goal Diff metric on Elite-GA; (c) #Avg. $\uparrow$ metric on BRKGA; (d) Goal Diff metric on BRKGA.
  • Figure 5: Performance comparison of MPI VS. its Variants across #FE budgets. The variants include No Gating, No Interpolation, and No Transfer. X-axis: #FEs; Y-axis: performance metrics. #Avg. $\uparrow$: the number of instances where the MPI's average objective value of the best solution (from 30 evaluations) exceeds the Variants'. Goal Diff: Win number minus Loss number from statistical testing (W-D-L results) on best solutions over 30 evaluations. Each subplot includes a red horizontal line dividing the area into Better (MPI outperforms the Variants) and Worse (the Variants outperforms MPI) areas. The subfigures present: (a) #Avg. $\uparrow$ metric on Elite-GA; (b) Goal Diff metric on Elite-GA; (c) #Avg. $\uparrow$ metric on BRKGA; (d) Goal Diff metric on BRKGA.
  • ...and 1 more figures