Hadjer Benmeziane

Hadjer Benmeziane

PhD Candidate

LAMIH, UPHF

About Me

I am a 2nd year PhD Candidate at LAMIH, Université Polytechnique des Hauts-de-France, supervised by Pr Niar Smail, Dr Ouarnoughi Hamza and Dr El Maghraoui Kaoutar (IBM Researcher).
My research vision is to enable deep learning powerful algorithms on tiny and resource-constrained devices. To achieve this, my work focus on developing an efficient, fast and task-less Neural Architecture Search method to automatically design deep neural networks.

During my master years, I founded School of AI Algiers. School of AI Algiers is a community of AI passionates that share their knowledge through organizing workshops and talks. Each week, I host a reading session 😀 ! Join us on this discord workspace.

Publications

A Comprehensive Survey on Hardware-aware Neural Architecture Search

Hadjer Benmeziane, Kaoutar El Maghraoui, Hamza Ouarnoughi, Smail Niar, Martin Wistuba, Naigang Wang

Survey

22/01/2021

We conduct a comprehensive survey on hardware-aware neural architecture search, present a new taxonomy and explore the components (i.e., search space, search algorithms and evaluation methods) and their limitations.

Paper

Hardware-aware Neural Architecture Search: Survey and Taxonomy

Hadjer Benmeziane, Kaoutar El Maghraoui, Hamza Ouarnoughi, Smail Niar, Martin Wistuba, Naigang Wang

Survey

04/2021

IJCAI 2021

In this survey, we focus on the hardware considerations in HW-NAS. Specifically, we explore which evaluation method is the best to characterize the latency, which search space is more hardware-friendly and which search exploration method is more flexible.

Paper Poster

Accelerating Neural Architecture Search with Rank-Preserving Surrogate Models

Hadjer Benmeziane, Hamza Ouarnoughi, Kaoutar El Maghraoui, Smail Niar

RS-NAS

07/2021

ArabWic 2021 - Best Student Paper Award

We present a new pairwise loss to train a surrogate model to accurately rank the deep learning architectures. We use this surrogate model to accelerate NAS algorithms from evolutionary algorithms to reinforcement learning.

Paper Presentation Code

Awards

WISC Scholarship

IEEE International Symposium on Women in Services Computing (WISC) 2021 Award

Best Student Paper Award

"RS-NAS" is awarded Best Student Paper Award at 7th Annual International Arab Women in Computing and Technology Conference.

WTM Scholar 2019/2020

Selected among 20 tech women in Europe and Africa to be awarded the Anita Borg's Scholarship.

Udacity Pytorch Side Project Winner -
Category: Cyber Security

with a Malware Classification Model created using ConvNets