Autoencoder-Based Generation of Individuals in Population-Based Metaheuristics

EasyChair Preprint 2650

Abstract

In this work, we explore a new way of breeding individuals in a population-based metaheuristic, by exploiting the ability of autoencoders to interpolate representation of candidate solutions in a learned latent space. In the remaining of this document, we present in more details our idea and show its application to the traveling salesman problem.

Keyphrases: Autoencoders, Metaheuristics, Optimization, Traveling Salesman Problem

@booklet{EasyChair:2650,
  author    = {Bastien Pasdeloup and Maryam Karimi-Mamaghan and Mehrdad Mohammadi and Patrick Meyer},
  title     = {Autoencoder-Based Generation of Individuals in Population-Based Metaheuristics},
  howpublished = {EasyChair Preprint 2650},
  year      = {EasyChair, 2020}}