Ihsen ALOUANI, PhD

News:

• A new paper accepted in CVPR 2024 : "DAP: A Dynamic Adversarial Patch for Evading Person Detectors", You can find it Here

• A new paper accepted in CVPR 2023 : "Jedi: Entropy-based Localization and Removal of Adversarial Patches", we propose a Shannon's entropy-based defense against physical adversarial patches. Key findings:

- We establish that adversarial patches have inherently high entropy distribution compared to benign images (including naturalistic patches)

- We leverage this observation to detect, localise and neutralise adversarial patches, recovering initial models performance

- We proposed a new adaptive attack that optimises the patch generation with an entropy budget; The generated patches evade detection BUT are no longer adversarial ... This lose-lose game is in favour of the defender.

A demo is available on Youtube: https://www.youtube.com/watch?v=8B5Mn9t7zc8&t=2s&ab_channel=IhsenAlouani

• EdgeAI project has started with an exciting Kick-off meeting in Catania, Italy. Check out the Project website : https://edge-ai-tech.eu/ 

• 5 PhD proposals to start fall 2023 in CSIT: interested in a PhD in a vibrant environment at Queen's University Belfast? You like to investigate hot topics in the security and privacy of machine learning ? Apply following the links here: https://www.linkedin.com/feed/update/urn:li:activity:7008384301431988225/ 

• New folks joining!  Two incredible Postdocs will be joining my team early 2023 Dr. Mahmoud Ghorbel and Imane Taibi ! Mahmoud is supported by RESIST project, and Imane is supported by EdgeAI. They will both investigate trustworthiness of Machine Learning. Welcome Mahmoud, Imane ! 

• Glad to announce that I recently joined Queen's University Belfast as a Senior Lecturer -- I'll be working with The Centre for Secure Information Technologies (CSIT) 

• New accepted paper at IJCNN 2022, Padua: "ROOM: Adversarial Machine Learning Attacks Under Real-Time Constraints" -- Check it out here 👉: IJCNN22. 

• Our paper entitled " Lower Voltage for Higher Security: Using Voltage Overscaling to Secure Deep Neural Networks" has been accepted for ICCAD 2021 ! 

• The project "RESIST: Robustness and Ethics of Intelligent Surveillance Systems" has been selected for funding ! -- We will be exploring Security, privacy and Ethics of Machine Learning applied to Video-surveillance systems. The kick-off is planned for November 2021  

• Our Chair proposal Luxant-ANVI : Neuromorphic Architectures for VIdeoprotection has been accepted for funding with a global budget of ~1.5 M€. The Chair is led by Prof. Pierre Boulet, co-funded by European Metropole of Lille, ISITE and Luxant Group. ( Fully funded PhD positions are open, feel free to reach out if interested! )

• Our paper entitled " Defensive Approximation: Securing CNNs using Approximate Computing" has been accepted ay ASPLOS 2021 conference! 

• Accepted paper in DATE 2021: "Securing Deep Spiking Neural Networks against Adversarial Attacks through Inherent Structural Parameters" R. El-Allami, A. Marchisio, M.  Shafique and I. Alouani

• Venu THATI, joined my team in Spetember 2020 for 2-year Postdoc to work on secure machine learning for Lab on Chip.  Venu got his PhD from KU Leuven, and we will be collaborating with Prof Julien Carlier (Physics Department of IEMN). Welcome Venu!

• Accepted paper in IEEE Sensors Journal (Q1): I. Jegham, A. Ben Khalifa, I. Alouani, and M. Mahjoub " Soft Spatial Attention-based Multimodal Driver Action Recognition Using Deep Learning" DOI: 10.1109/JSEN.2020.3019258

• Accepted paper in IJCNN2020 The International Joint Conference on Neural Networks: V. Venceslai, A. Marchisio, I. Alouani, M. Martina and M. Shafique "NeuroAttack: Undermining Spiking Neural Networks Security through Externally Triggered Bit-Flips" (ArXiv link: https://arxiv.org/abs/2005.08041 )

• Accepted WiP paper in DAC2020:  Md S Islam, I. Alouani, K. Khasawneh "Approx-HMDs: Approximating Hardware MalwareDetectors to Defend Against Adversarial Attacks"

• A new accepted paper in MDPI Electronics "OLIMP: A Heterogeneous Multimodal dataset for Advanced Environment Perception".  OLIMP presents a new multimodal dataset for Intelligent Transport Systems. To download the dataset and read the paper, please check the Datasets page.