About me

I am a PostDoc at the Vision Lab in University of California, Los Angeles (UCLA). I obtained my PhD from the signal processing laboratory in EPFL in 2016, and my M.Sc. in Electrical Engineering from EPFL in 2012. I was awarded the IBM PhD Fellowship awards for the academic years 2013-2014 and 2015-2016.

Research interests: I am broadly interested in challenging problems related to computer vision and machine learning. In my recent research, I have focused on analyzing the robustness and invariance of classifiers to transformations from an empirical and theoretical perspective.
Curriculum Vitae
Email: fawzi [AT] cs.ucla [dot] edu


Publications

2017


Universal adversarial perturbations
Seyed-Mohsen Moosavi-Dezfooli*, Alhussein Fawzi*, Omar Fawzi, Pascal Frossard (*: Equal contribution)
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, Hawaii, 2017. [Oral presentation]

Slides
Check the demo on YouTube!
Code available on GitHub

2016


Robustness of classifiers: from adversarial to random noise
Alhussein Fawzi*, Seyed-Mohsen Moosavi-Dezfooli*, Pascal Frossard (*: Equal Contribution)
Neural Information Processing Systems (NIPS), 2016.

DeepFool: a simple and accurate method to fool deep neural networks
Seyed-Mohsen Moosavi-Dezfooli, Alhussein Fawzi, Pascal Frossard
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, USA, 2016.
Code

Measuring the effect of nuisance variables on classifiers
Alhussein Fawzi, Pascal Frossard
British Machine Vision Conference (BMVC), York, UK, 2016. [Oral presentation]

Multi-task additive models with shared transfer functions
Alhussein Fawzi, Mathieu Sinn, Pascal Frossard
IEEE Transactions on Signal Processing, 2016.
Version presented at the ICML Workshop on Demand Forecasting and Machine Learning, Lille, France, 2015.

Structured Dimensionality Reduction for Additive Model Regression
Alhussein Fawzi, Jean-Baptiste Fiot, Bei Chen, Mathieu Sinn, Pascal Frossard
IEEE Transactions on Knowledge and Data Engineering (TKDE), 2016.

Adaptive data augmentation for image classification
Alhussein Fawzi, Horst Samulowitz, Deepak Turaga, Pascal Frossard
International Conference on Image Processing (ICIP), Phoenix, Arizona, 2016.

2015


Analysis of classifiers' robustness to adversarial perturbations
Alhussein Fawzi, Omar Fawzi, Pascal Frossard
Arxiv pre-print arXiv:1502.02590

Dictionary learning for fast classification based on soft-thresholding
Alhussein Fawzi, Mike Davies, Pascal Frossard
International Journal of Computer Vision (IJCV), 2015. [Official version]
MATLAB code

Manitest: Are classifiers really invariant?
Alhussein Fawzi, Pascal Frossard
British Machine Vision Conference (BMVC), Swansea, UK, 2015.
Project page

Fundamental limits on adversarial robustness
Alhussein Fawzi, Omar Fawzi, Pascal Frossard
ICML Workshop on Deep Learning, Lille, France, 2015.

2014 and earlier


Image registration with sparse approximations in parametric dictionaries
Alhussein Fawzi, Pascal Frossard
SIAM Journal on Imaging Sciences, 2013. [Official version]
Technical report

Classification of unions of subspaces with sparse representations
Alhussein Fawzi, Pascal Frossard
Asilomar Conference on Signals, Systems and Computers, 2013.

A geometric framework for registration of sparse images
Alhussein Fawzi, Pascal Frossard
IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2013.
Version presented in SPARS 2013.

Thresholding-based reconstruction of compressed correlated signals
Alhussein Fawzi, Tamara Tosic, Pascal Frossard
IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2012.
Technical report.

Theses


Geometric group sparsity in image analysis
Alhussein Fawzi
Master's thesis, EPFL.

Software

Universal adversarial perturbations

Implementation of the algorithm for computing universal perturbations
GitHub page

DeepFool

Implementation of the DeepFool algorithm for fooling deep neural networks
GitHub page

Manitest

MATLAB and C++ (with OpenCV) implementations of Manitest to compute the invariance of classifiers to geometric transformations.
See project webpage for download and more information.

Learning Algorithm for Soft-Thresholding (LAST)

Implementation of the DC-based dictionary learning algorithm for soft-thresholding based based classifiers.
Download MATLAB code.