An algorithmic framework for the optimization of deep neural networks architectures and hyperparameters

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Keisler, Julie, Talbi, El-Ghazali, Claudel, Sandra, Cabriel, Gilles
Format: Preprint
Veröffentlicht: 2023
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866916244508114944
author Keisler, Julie
Talbi, El-Ghazali
Claudel, Sandra
Cabriel, Gilles
author_facet Keisler, Julie
Talbi, El-Ghazali
Claudel, Sandra
Cabriel, Gilles
contents In this paper, we propose an algorithmic framework to automatically generate efficient deep neural networks and optimize their associated hyperparameters. The framework is based on evolving directed acyclic graphs (DAGs), defining a more flexible search space than the existing ones in the literature. It allows mixtures of different classical operations: convolutions, recurrences and dense layers, but also more newfangled operations such as self-attention. Based on this search space we propose neighbourhood and evolution search operators to optimize both the architecture and hyper-parameters of our networks. These search operators can be used with any metaheuristic capable of handling mixed search spaces. We tested our algorithmic framework with an evolutionary algorithm on a time series prediction benchmark. The results demonstrate that our framework was able to find models outperforming the established baseline on numerous datasets.
format Preprint
id arxiv_https___arxiv_org_abs_2303_12797
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle An algorithmic framework for the optimization of deep neural networks architectures and hyperparameters
Keisler, Julie
Talbi, El-Ghazali
Claudel, Sandra
Cabriel, Gilles
Neural and Evolutionary Computing
Artificial Intelligence
Machine Learning
In this paper, we propose an algorithmic framework to automatically generate efficient deep neural networks and optimize their associated hyperparameters. The framework is based on evolving directed acyclic graphs (DAGs), defining a more flexible search space than the existing ones in the literature. It allows mixtures of different classical operations: convolutions, recurrences and dense layers, but also more newfangled operations such as self-attention. Based on this search space we propose neighbourhood and evolution search operators to optimize both the architecture and hyper-parameters of our networks. These search operators can be used with any metaheuristic capable of handling mixed search spaces. We tested our algorithmic framework with an evolutionary algorithm on a time series prediction benchmark. The results demonstrate that our framework was able to find models outperforming the established baseline on numerous datasets.
title An algorithmic framework for the optimization of deep neural networks architectures and hyperparameters
topic Neural and Evolutionary Computing
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2303.12797