ETRO VUB
About ETRO  |  News  |  Events  |  Vacancies  |  Contact  
Home Research Education Industry Publications About ETRO

ETRO Publications

Full Details

Conference Publication

Neural Architecture Search under black-box objectives with deep reinforcement learning and increasingly-sparse rewards

Host Publication: International Conference on Artificial Intelligence in Information and Communication

Authors: M. Perez Gonzalez, A. Díaz Berenguer, M. Oveneke and H. Sahli

Publisher: IEEE

Publication Date: Feb. 2020

ISBN: 978-1-7281-4986-8


Abstract:

In this paper, we address the problem of neural architecture search (NAS) in a context where the optimality policy derivatives. In this scenario, O(A) typically provides readings from a set of sensors on how a neural network architecture A fares in a target hardware, including its: power consumption, working temperature, cpu/gpu usage, central bus occupancy, and more. Current differentiable NAS approaches fail in this problem context due to lack of access to derivatives, whereas traditional reinforcement learning NAS approaches remain too expensive computationally. As solution, we propose a reinforcement learning NAS strategy based on policy gradient with increasingly sparse rewards. We rely on the fact that one does not need to fully train the weights of two neural networks to compare them. Our solution starts by comparing architecture candidates with almost fixed weights and no training, and progressively shifts toward comparisons under full weights training. Experimental results confirmed both the accuracy and training efficiency of our solution, as well as its compliance with soft/hard constraints imposed on the sensors feedback. Our strategy allows finding near-optimal architectures significantly faster, in approximately 1/3 of the time it would take otherwise.

Other Reference Styles
Current ETRO Authors

Dr. Mitchel Perez Gonzalez

+32 (0)02 629 298

maperezg@etrovub.be

more info

Mr. Abel Díaz Berenguer

+32 (0)02 629 102

aberengu@etrovub.be

more info

Prof. Hichem Sahli

+32 (0)02 629 291

hsahli@etrovub.be

more info

Other Publications

• Journal publications

IRIS • LAMI • AVSP

• Conference publications

IRIS • LAMI • AVSP

• Book publications

IRIS • LAMI • AVSP

• Reports

IRIS • LAMI • AVSP

• Laymen publications

IRIS • LAMI • AVSP

• PhD Theses

Search ETRO Publications

Author:

Keyword:  

Type:








- Contact person

- IRIS

- AVSP

- LAMI

- Contact person

- Thesis proposals

- ETRO Courses

- Contact person

- Spin-offs

- Know How

- Journals

- Conferences

- Books

- Vacancies

- News

- Events

- Press

Contact

ETRO Department

info@etro.vub.ac.be

Tel: +32 2 629 29 30

©2020 • Vrije Universiteit Brussel • ETRO Dept. • Pleinlaan 2 • 1050 Brussels • Tel: +32 2 629 2930 (secretariat) • Fax: +32 2 629 2883 • WebmasterDisclaimer