Permutation-based optimization using a generative adversarial network
Sami Lemtenneche, Abdelhakim Cheriet, Bensayah Abdellah · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2021
This paper presents a new method to deal with permutation-based problems using a Generative Adversarial Network. Generative adversarial Networks (GANs) are used in many fields as models that mimic the trained data distribution. However, dealing with permutation space in GANs has not been yet investigated in detail. To reach this objective, we propose using GANs to solve the permutation-based optimization problem using an Estimation of Distribution Algorithm. We use a Random Key method to translate the generated data from continuous values to permutations. The Experiment aims to investigate the quality of solutions provided by the proposed method in two ways. Firstly, we evaluate the solutions obtained directly from the GANs, and in the second one, we use a local search method to improve the quality of the solution. In the end, We perform a set of experiments on instances of the known TSP problems. Also, we notice that the obtained results promise to extend the work on other classes of problems.