Motasem Alfarra

Machine Learning Researcher at Qualcomm AI Research, Amsterdam, Netherlands

Qualcomm AI Research

Biography

I am a machine learning researcher at Qualcomm AI Research in Amsterdam, Netherlands. I obtained my Ph.D. in Electrical and Computer Engineering from KAUST in Saudi Arabia advised by Prof. Bernard Ghanem. I also obtained my M.Sc degree in Electrical Engineering from KAUST, and my undergraduate degree in Electrical Engineering from Kuwait University. I am interested in domain shifts, LLM safety, and how to combat them with test-time adaptation and continual learning. I helped co-organizing the first workshop on Test-Time Adaptation at CVPR2024!

Download my resumé.

Interests

LLM Safety
Test Time Adaptation
Continual Learning
Computer Vision and Machine Learning

Education

PhD in Electrical and Computer Engineering, 2020-2024
KAUST
MSc in Electrical and Computer Engineering, 2019-2020
KAUST
BSc in Electrical Engineering, 2014-2018
Kuwait University

Experience

Machine Learning Researcher

Qualcomm AI Research

Sep 2024 – Present Amsterdam, Netherlands

Research Scienctist at Qualcomm AI Research, Amsterdam.

Research Internship

Qualcomm AI Research

May 2023 – Sep 2023 Amsterdam, Netherlands

Research internship at the distributed learning team in Qualcomm AI Research.

Research Internship

Intel

Aug 2022 – Jan 2023 Münich, Germany

Research internship at the Embodied AI Lab at Intel supervised by Matthias Müller.

Research Internship

University of Oxford

Oct 2021 – Feb 2022 Oxford, United Kingdom

Research visit to the Torr Vision Group (TVG) supervised by Prof. Philip Torr.

Teacher Assistant

KAUST

Jan 2021 – May 2021 Saudi Arabia

I was a TA for the Ph.D course “Introduction to Computer Vision”.

Guest Lecturer

Universidad Panamericana

Jul 2020 – Jul 2020 Mexico

I gave one lecture titled “Adversarial Attacks and Network Robustness”.

Research Assistant

Kuwait University

Jul 2017 – May 2018 Kuwait

Developed algorithms and implemented network simulations.

Featured Publications

Test-Time Adaptation with Source Based Auxiliary Tasks

We leverage source-based auxiliary tasks to accelerate and enhance adaptation.

Motasem Alfarra, Alvaro Correia, Bernard Ghanem, Christos Louizos

Test-Time Adaptation with Source Based Auxiliary Tasks

Evaluation of Test-Time Adaptation Under Computational Time Constraints

We propose a novel online evaluation protocol for Test Time Adaptation (TTA) methods, which penalizes slower methods by providing them with fewer samples for adaptation.

Motasem Alfarra, Hani Itani, Alejandro Pardo, Shyma Alhuwaider, Merey Ramazanova, Juan C. Pérez, Zhipeng Cai, Matthias Müller, Bernard Ghanem

Evaluation of Test-Time Adaptation Under Computational Time Constraints

SimCS: Simulation for Domain Incremental Online Continual Segmentation

We leverage simulated data to mitigate forgetting in domain incremental continual segmentation.

Motasem Alfarra, Zhipeng Cai, Adel Bibi, Bernard Ghanem, Matthias Müller

SimCS: Simulation for Domain Incremental Online Continual Segmentation

ANCER: Anisotropic Certification via Sample-wise Volume Maximization

We extend randomized smoothing to anisotropic distributions and optimize the anisotropic smoothing parameters per input point in randomized smoothing.

Motasem Alfarra, Francisco Eiras, M. Pawan Kumar, Philip H. S. Torr, Puneet K. Dokania, Bernard Ghanem, Adel Bibi

ANCER: Anisotropic Certification via Sample-wise Volume Maximization

On the Robustness of Quality Measures for GANs

We assess the adversarial robustness of Inception Score and Frechet Inception Distance (FID) and propose a robustified version of FID.

Motasem Alfarra, Juan C. Pérez, Anna Frühstück, Philip H. S. Torr, Peter Wonka, Bernard Ghanem

On the Robustness of Quality Measures for GANs

3DeformRS: Certifying Spatial Deformations on Point Clouds

We design the anti-adversary layer that enhances the robustness of pretrained models against strong adversarial attacks.

Motasem Alfarra, Gabriel Pérez, Juan C. Pérez, Silvio Giancola, Bernard Ghanem

3DeformRS: Certifying Spatial Deformations on Point Clouds

Combating Adversaries with Anti-Adversaries

We design the anti-adversary layer that enhances the robustness of pretrained models against strong adversarial attacks.

Motasem Alfarra, Juan C. Pérez, Ali Thabet, Adel Bibi, Philip H. S. Torr, Bernard Ghanem

Combating Adversaries with Anti-Adversaries

DeformRS: Certifying Input Deformations with Randomized Smoothing

We extend randomized smoothing to certify image deformations such as rotation, translation, scaling, and affine.

Motasem Alfarra, Adel Bibi, Naeemullah Khan, Philip H. S. Torr, Bernard Ghanem

DeformRS: Certifying Input Deformations with Randomized Smoothing

Gabor Layers Enhance Network Robustness

We replace the first convolutional layer in deep neural networks with a Gabor layer to enhance networks robustness.

Motasem Alfarra, Juan C. Pérez, Guillaume Jeanneret, Adel Bibi, Ali Thabet, Bernard Ghanem, Pablo Arbeláez

On the Decision Boundaries of Neural Networks: A Tropical Geometry Perspective

We characterizes the decision boundaries of neural networks using tropical geometry, and leverage this characterization into several applications.

Motasem Alfarra, Adel Bibi, Hasan Hammoud, Mohamed Gaafar, Bernard Ghanem

On the Decision Boundaries of Neural Networks: A Tropical Geometry Perspective

Recent Publications

Quickly discover relevant content by filtering publications.

Merey Ramazanova, Alejandro Pardo, Bernard Ghanem, Motasem Alfarra (2025). Combating Missing Modalities in Egocentric Videos at Test Time. In ICLR.

PDF Cite

Kumail Alhamoud, Yasir Ghunaim, Motasem Alfarra, Thomas Hartvigsen, Philip H. S. Torr, Bernard Ghanem, Adel Bibi, Marzyeh Ghassemi (2024). Generalizability of Adversarial Robustness Under Distribution Shifts. In MICCAI.

PDF Cite Code Poster

Kumail Alhamoud, Hasan Abed Al Kader Hammoud, Motasem Alfarra, Bernard Ghanem (2023). Generalizability of Adversarial Robustness Under Distribution Shifts. In TMLR [Featured].

PDF Cite

Joachim Houyon, Anthony Cioppa, Yasir Ghunaim, Motasem Alfarra, Anais Halin, Maxim Henry, Bernard Ghanem, Marc Van Droogenbroeck (2023). Online Distillation with Continual Learning for Cyclic Domain Shifts. In CVPRW'23.

PDF Cite Code

Juan C. Pérez, Motasem Alfarra, Ali Thabet, Pablo Arbeláez, Bernard Ghanem (2023). Towards Assessing and Characterizing the Semantic Robustness of Face Recognition. In CVPRW'23.

PDF Cite

See all publications

Contact

motasem.alfarra@kaust.edu.sa, malfarra@qti.qualcomm.com
Science Park, Amsterdam, 1098 XH
Matrix 1 Building - Science Park, Amsterdam, Netherlands