Conference Proceeding

A NEAT Visualisation of Neuroevolution Trajectories

Citation

Sarti S & Ochoa G (2021) A NEAT Visualisation of Neuroevolution Trajectories. In: Castillo PA & Jiménez Laredo JL (eds.) Applications of Evolutionary Computation. Lecture Notes in Computer Science, 12694. 24th International Conference, EvoApplications 2021, Seville, Spain, 07.04.2021-09.04.2021. Cham, Switzerland: Springer, pp. 714-728. https://doi.org/10.1007/978-3-030-72699-7_45

Abstract
NeuroEvolution of Augmenting Topologies (NEAT) is a system for evolving neural network topologies along with weights that has proven highly effective and adaptable for solving challenging reinforcement learning tasks. This paper analyses NEAT through the lens of Search Trajectory Networks (STNs), a recently proposed visual approach to study the dynamics of evolutionary algorithms. Our goal is to improve the understanding of neuroevolution systems. We present a visual and statistical analysis contrasting the behaviour of NEAT, with and without using the crossover operator, when solving the two benchmark problems outlined in the original NEAT article: XOR and double-pole balancing. Contrary to what is reported in the original NEAT article, our experiments without crossover perform significantly better in both domains.

Keywords
Neuroevoltuion; NEAT; Search Trajectory Networks

StatusPublished
Title of seriesLecture Notes in Computer Science
Number in series12694
Publication date01/04/2021
Publication date online01/04/2021
URLhttp://hdl.handle.net/1893/32649
PublisherSpringer
Place of publicationCham, Switzerland
ISSN of series0302-9743
ISBN978-3-030-72698-0
eISBN978-3-030-72699-7
Conference24th International Conference, EvoApplications 2021
Conference locationSeville, Spain
Dates