Youssef Tamaazousti, PhD
Postdoctoral Research Associate @MIT and @HBKU
Computer Science and Artificial Intelligence Laboratory (CSAIL)
and Qatar Computing Research Institute (QCRI)
ytamaaz[at]mit[dot]edu

About Me

I am currently a Postdoctoral research associate at Computer Science and Artificial Intelligence Laboratory (CSAIL), MIT and Qatar Computing Research Institute (QCRI), HBKU. I am working with Professor Antonio Torralba, Dr. Ferda Ofli and Dr. Ingmar Weber. My research lies at the intersection of computer vision, machine learning and multimedia modeling. Specifically, I am interested in building food recognition systems and applying them on social media data in order to understand health habits.
Previously (in June 2018), I obtained a PhD in Computer-Vision and Machine-Learning at CentraleSupélec and Alternative Energies & Atomic Energy Commission (CEA), under the supervision of Dr. Hervé Le Borgne (CEA) and Professor Céline Hudelot (Centrale-Supélec).



Research

My research centers around Deep-Learning and its applications to Computer-Vision, in particular object/scene recognition, content-based image retrieval and video recognition. More specifically, I am interested in the learning of image representations using Deep Convolutional Neural Networks (DCNNs) in order to re-use them (through a transfer-learning scheme) when facing with small datasets that contains too few images and/or classes to learn representations from scratch. The learning of these image-representations is based on the unsupervised generative aspect of DCNNs which is quite powerful but complex, making it hard to understand and thus very hard to control as we want.
Keywords: Computer-Vision, Machine-Learning, Deep-Learning, Image Understanding, Object/Scene Recognition.



Publications

2018

Learning Finer-class Networks for Universal Representations


Julien Girard*, Youssef Tamaazousti*, Hervé Le Borgne, and Céline Hudelot
*Both authors contributed equally.
BMVC 2018 British Machine Vision Conference, Newcastle, UK, September 2018
Acceptance rate: 29.9% (258/862)
PDF Supplementary BibTex

On The Universality of Visual and Multimodal Representations


Youssef Tamaazousti
PhD Thesis at CentraleSupélec and CEA LIST, Defended on June 1st 2018
Comitee: Mathieu Cord (UPMC), Philippe-Henri Gosselin (Technicolor), Iasonas Kokkinos (Facebook and UCL), Florent Perronnin (NaverLabs), Pablo Piantanida (CentraleSupélec)
PDF BibTex Slides

Learning More Universal Representations for Transfer-Learning


Youssef Tamaazousti, Hervé Le Borgne, Céline Hudelot, Mohamed El Amine Seddik and Mohamed Tamaazousti
PAMI 2018 Journal of IEEE Transactions on Pattern Analysis and Machine Intelligence (submitted)
(under review)
PDF Supplementary BibTex Code (Github)

Procédé d’obtention d’apprentissage d’un premier réseau de neurones convolutif vers un deuxième réseau de neurones convolutif


Youssef Tamaazousti, Julien Girard, Hervé Le Borgne and Céline Hudelot
Patent 2018 Filed June 2018


2017

Vision-Language Integration using Constrained Local Semantic Features


Youssef Tamaazousti, Hervé Le Borgne, Adrian Popescu, Etienne Gadeski, Alexandru Lucian Ginsca, and Céline Hudelot
CVIU 2017 Journal of Computer Vision and Image Understanding (In press)
Journal - Impact Factor: 2.49
PDF BibTex

MuCaLe-Net: Multi Categorical-Level Networks to Generate More Discriminating Features


Youssef Tamaazousti, Hervé Le Borgne, and Céline Hudelot
CVPR 2017 Computer Vision and Pattern Recognition, Hawaii, USA, July 2017
Acceptance rate: 29% (783/2680)
PDF BibTex Supplementary Poster Code (Github)

AMECON: Abstract Meta Concept Features for Text-Illustration


Ines Chami*, Youssef Tamaazousti* and Hervé Le Borgne
*Both authors contributed equally.
ICMR 2017 International Conference on Multimedia Retrieval, Bucharest, Romania, June 2017
Oral - Acceptance rate: 21%
PDF BibTex Slides Poster Code (coming soon)

Supervised Learning of Entity Disambiguation Models by Negative Sample Selection


Hani Daher, Romaric Besançon, Olivier Ferret, Hervé Le Borgne, Anne-Laure Daquo, and Youssef Tamaazousti
CICLing 2017 International Conference on Computational Linguistics and Intelligent Text Processing, Budapest, Hungary, April 2017
PDF BibTex

Déscripteur Sémantique Local Contraint Basé sur un RNC Diversifié


Youssef Tamaazousti, Hervé Le Borgne, Adrian Popescu, Etienne Gadeski, Alexandru Lucian Ginsca, and Céline Hudelot
TS 2017 Journal of Traitement du Signal (In press)
Journal
PDF BibTex

Désambiguïsation d'entités nommées par apprentissage de modèles d'entités à large échelle


Hani Daher, Romaric Besançon, Olivier Ferret, Hervé Le Borgne, Anne-Laure Daquo, and Youssef Tamaazousti
CORIA 2017 COnférence en Recherche d'Information et Applications, Marseille, France, March 2017
PDF BibTex

2016

Procédé d’obtention d’un système de labellisation d’images


Youssef Tamaazousti, Hervé Le Borgne and Céline Hudelot
Patent 2016 Filed September 2016


Diverse Concept-Level Features for Multi-Object Classification


Youssef Tamaazousti, Hervé Le Borgne and Céline Hudelot
ICMR 2016 International Conference on Multimedia Retrieval, New York, USA, June 2016
Oral - Acceptance rate: 18%

PDF BibTex Slides

Constrained Local Enhancement of Semantic Features by Content-Based Sparsity


Youssef Tamaazousti, Hervé Le Borgne and Adrian Popescu
ICMR 2016 International Conference on Multimedia Retrieval, New York, USA, June 2016
Oral - Acceptance rate: 18%

PDF BibTex Slides

Image Annotation and Two Paths to Text Illustration


Hervé Le Borgne, Etienne Gadeski, Ines Chami, Thi Quynh Nhi Tran, Youssef Tamaazousti, Alexandru Lucian Ginsca, and Adrian Popescu
ImageCLEF 2016 CLEF 2016 Evaluation Labs and Workshop, Online Working Notes, Evora, Portugal, Semptember 2016

PDF BibTex

Descripteurs à divers niveaux de concepts pour la classification d’images multi-objets


Youssef Tamaazousti, Hervé Le Borgne and Céline Hudelot
RFIA 2016 Reconnaissance des Formes et Intelligence Artificielle, Clermont Ferrand, France, June 2016
Oral - Acceptance rate: 39%

PDF BibTex Slides

Agrégation de descripteurs sémantiques locaux contraints par parcimonie basée sur le contenu


Youssef Tamaazousti, Hervé Le Borgne and Adrian Popescu
RFIA 2016 Reconnaissance des Formes et Intelligence Artificielle, Clermont Ferrand, France, June 2016
Oral - Acceptance rate: 39%

PDF BibTex Slides

Teaching / Supervising

Supervising: MSc students



- Julien Girard, @CEA LIST, with Hervé Le Borgne, in 2018 (will start a PhD at CEA LIST)
- Inès Chami, @CEA LIST, with Hervé Le Borgne, in 2017 (will start a PhD at Stanford University)




Practical Artificial-Intelligence at Centrale-Supélec


Artificial Intelligence tutorials for Master of Science (M.Sc) students.
Rational, Expert and Logical agents, Machine Learning, Basics of NLP, Basics of Computer-Vision, etc.
Framework: Keras. [Python]

Youssef Tamaazousti with Céline Hudelot
Centrale-Supélec - 2017

Practical Deep-Learning at Centrale-Supélec


Deep Learning tutorials for Master of Science (M.Sc) students.
Linear Regression, Multi-Layer Perceptron (MLP), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), Transfer-Learning, Fine-Tuning, etc.
Framework: TensorFlow. [Python]

Youssef Tamaazousti with Hervé Le Borgne
Centrale-Supélec - 2017

Practical Human-Computer Interaction


Human-Computer Interaction for 1st year IUT students.

Youssef Tamaazousti
IUT d'Orsay (University of Paris-Sud) - 2017

Reviewing Activities

ACCV 2018


Asian Conference on Computer Vision 2018: Two papers.




CVPR 2018


IEEE Computer Vision and Pattern Recognition 2018: Seven papers.




ICLR 2018


International Conference on Learning Representations 2018: Two papers.




NIPS 2017


Advances in Neural Information Processing Systems 2017: Two papers.




Projects

AI4DTY website


A website dedicated to novice students (of Centrale-Supelec) in the field of Artificial Intelligence (AI), that lists many useful links, including theoretical and practical courses, frameworks, databases and misc tools.

Youssef Tamaazousti
Project for Centrale-Supélec
Link to AI4DTY website

Pedestrian Detection


Development of a module of a driving assistance application. The module consisted to detect pedestrians on video-frames obtained from an embedded camera. I developed it using OpenCV and LibSVM libraries. [C++]

Youssef Tamaazousti under the supervision of Florence Rossant
Scholar Project

Road Signs Detection and Recognition


Development of a module of a driving assistance application. The module consisted to detect and recognize road signs on video-frames obtained from an embedded camera. I developed it using OpenCV and LibSVM libraries. [C++]

Youssef Tamaazousti under the supervision of Florence Rossant
Scholar Project

Hand-Written Digit Recognition


Development of a hand-written digit recognition system using a shallow neural network that contains multiple hidden-layers. I developed it from scratch, i.e, without any library or framework. [Matlab]

Youssef Tamaazousti under the supervision of Mathieu Manceny
Scholar Project

Curriculum vitae






Download my CV