EvoAAA: An evolutionary methodology for automated \neural autoencoder architecture search

TL;DR

This paper introduces EvoAAA, an evolutionary approach for automating the design of autoencoder neural network architectures, improving their efficiency and effectiveness in feature representation across diverse datasets.

Contribution

It presents a novel evolutionary methodology for automatic autoencoder architecture search, reducing manual effort and enhancing model performance.

Findings

EvoAAA outperforms manual design in architecture quality.

The approach finds compact, information-rich representations.

Results are consistent across nine diverse datasets.

Abstract

Machine learning models work better when curated features are provided to them. Feature engineering methods have been usually used as a preprocessing step to obtain or build a proper feature set. In late years, autoencoders (a specific type of symmetrical neural network) have been widely used to perform representation learning, proving their competitiveness against classical feature engineering algorithms. The main obstacle in the use of autoencoders is finding a good architecture, a process that most experts confront manually. An automated autoencoder architecture search procedure, based on evolutionary methods, is proposed in this paper. The methodology is tested against nine heterogeneous data sets. The obtained results show the ability of this approach to find better architectures, able to concentrate most of the useful information in a minimized coding, in a reduced time.