CNNGen: A Generator and a Dataset for Energy-Aware Neural Architecture Search

Antoine Gratia; Hong Liu; Shin’ichi Satoh; Paul Temple; Pierre-Yves Schobbens; Gilles Perrouin

CNNGen: A Generator and a Dataset for Energy-Aware Neural Architecture Search

Antoine Gratia, Hong Liu, Shin’ichi Satoh, Paul Temple, Pierre-Yves Schobbens, Gilles Perrouin

Résultats de recherche: Contribution dans un livre/un catalogue/un rapport/dans les actes d'une conférence › Article dans les actes d'une conférence/un colloque

15 Téléchargements (Pure)

Résumé

Neural Architecture Search (NAS) methods seek optimal networks by exploring thousands of variants of a reference architecture. Yet, optimality is typically related to prediction performance, overlooking the environmental impacts of training. Thus, NAS search spaces are unfit for performance and energy consumption trade-offs. We contribute to energy-aware NAS with (i) a grammar-based Convolutional Neural Network generator (CNNGen) producing diverse architectures not based on a reference one; (ii) 1,300 available architectures obtained via CNNGen with their implementation, energy consumption and performance measurements; (iii) Three state-of-the-art predictors releasing the need for trained models for performance and energy estimation.

langue originale	Indéfini/inconnu
titre	32nd European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning
Etat de la publication	Publié - 2024

SDG des Nations Unies

Ce résultat contribue à ou aux Objectifs de développement durable suivants

Accès au document

ESANN2024_CNNGEN-7Accepted author manuscript, 402 KB

Contient cette citation

@inproceedings{8c003e55d6a743489d7c4b7b16e953c3,

title = "CNNGen: A Generator and a Dataset for Energy-Aware Neural Architecture Search",

abstract = "Neural Architecture Search (NAS) methods seek optimal networks by exploring thousands of variants of a reference architecture. Yet, optimality is typically related to prediction performance, overlooking the environmental impacts of training. Thus, NAS search spaces are unfit for performance and energy consumption trade-offs. We contribute to energy-aware NAS with (i) a grammar-based Convolutional Neural Network generator (CNNGen) producing diverse architectures not based on a reference one; (ii) 1,300 available architectures obtained via CNNGen with their implementation, energy consumption and performance measurements; (iii) Three state-of-the-art predictors releasing the need for trained models for performance and energy estimation.",

author = "Antoine Gratia and Hong Liu and Shin{\textquoteright}ichi Satoh and Paul Temple and Pierre-Yves Schobbens and Gilles Perrouin",

year = "2024",

language = "Ind{\'e}fini/inconnu",

isbn = "978-2-87587-090-2",

booktitle = "32nd European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning",

}

CNNGen: A Generator and a Dataset for Energy-Aware Neural Architecture Search. / Gratia, Antoine; Liu, Hong; Satoh, Shin’ichi et al.
32nd European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning. 2024.

Résultats de recherche: Contribution dans un livre/un catalogue/un rapport/dans les actes d'une conférence › Article dans les actes d'une conférence/un colloque

TY - GEN

T1 - CNNGen: A Generator and a Dataset for Energy-Aware Neural Architecture Search

AU - Gratia, Antoine

AU - Liu, Hong

AU - Satoh, Shin’ichi

AU - Temple, Paul

AU - Schobbens, Pierre-Yves

AU - Perrouin, Gilles

PY - 2024

Y1 - 2024

N2 - Neural Architecture Search (NAS) methods seek optimal networks by exploring thousands of variants of a reference architecture. Yet, optimality is typically related to prediction performance, overlooking the environmental impacts of training. Thus, NAS search spaces are unfit for performance and energy consumption trade-offs. We contribute to energy-aware NAS with (i) a grammar-based Convolutional Neural Network generator (CNNGen) producing diverse architectures not based on a reference one; (ii) 1,300 available architectures obtained via CNNGen with their implementation, energy consumption and performance measurements; (iii) Three state-of-the-art predictors releasing the need for trained models for performance and energy estimation.

AB - Neural Architecture Search (NAS) methods seek optimal networks by exploring thousands of variants of a reference architecture. Yet, optimality is typically related to prediction performance, overlooking the environmental impacts of training. Thus, NAS search spaces are unfit for performance and energy consumption trade-offs. We contribute to energy-aware NAS with (i) a grammar-based Convolutional Neural Network generator (CNNGen) producing diverse architectures not based on a reference one; (ii) 1,300 available architectures obtained via CNNGen with their implementation, energy consumption and performance measurements; (iii) Three state-of-the-art predictors releasing the need for trained models for performance and energy estimation.

M3 - Article dans les actes d'une conférence/un colloque

SN - 978-2-87587-090-2

BT - 32nd European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning

ER -