Aymen Hamrouni

Aymen Hamrouni

Graduate Student and Researcher; Deep Learning and Algorithm Design

Projects

Current Projects (2021-2022)

A Video-based Visual Transformer for Crowd Management Monitoring (Academic)

  • No further information could be given at the moment

Multi-modal Asymmetric Autoencoders for Massive Photo Collection Applications (Academic)

  • No further information could be given at the moment

Generative Adversarial Networks for de-noising Real-time Video feed Corrupted with Environmental Effects for Autonoums Navigation systems (Academic)

  • Creating a low-latency video processing pipeline where videos captured by a low-quality camera equipped to the navigation systems of a device (e.g., UAV, autonomous vehicle) is cleaned from rain droplets and rainstreaks and other environmental effects and then reconstructed and enhanced. The approaches relies on a hybrid deep learning architecture tweaked from auto-encoders and generative adversarial networks.

Team formation and Ressource Allocation in Collaborative Mobile Crowdsroucing for Social IoT (Academic)

  • Formulating and solving optimal team formation problems to create groups of IoT devices that match specific required tasks. The formulated approach serves also as community detection and resource allocation for different components in the IoT network. As the problem is NP-hard, heuristic Fuzzy-logic approaches are proposed.

UAVs for Crowd-management and Navigability in Social IoT (Academic)

  • Enabling crowd-management systems using UAVs by designing multi-UAV fleet coordination, path planning and navigation, and object detection and recognition while taking into consideration privacy-aware and anti-intrusion challenges.

Previous Projects (2019-2021)

Low-complexity Graph Neural Networks for Socially Connected IoT systems (Academic)

  • Designed low-complexity meta-heuristic algorithms for team formation and recruitment in collaborative mobile crowdsoucing using social Internet-of-Things networks. Enabled innovative Graph Neural Network techniques for service discovery in social IoT.

Social IoT for Collaborative Mobile Crowdsourcing on Facebook (Industry)

  • No further information could be given at the moment

Mobile Crowdsourcing Image-based Event Reporting System (Academic)

  • Designed two heuristic low-latency AI-powered redundancy filtering and quality check system for captured images in Mobile Crowdsourcing frameworks. After the IoT devices capture the needed photo of the event, wrong, inaccurate, and redundant images must be dismissed before being uploaded to save resources such as energy and bandwidth. The first designed approach relied on an optimized and compressed Convolutional Neural Network (CNN) with a graph data-structure P-tree search while the second approach relied on a mixed and hierarchical multi-modal auto-encoder with meta-data embedding and clustering analysis.

Recruitment and Scheduling in Spatial Mobile Crowdsroucing (Academic)

  • Designed several deterministic algorithms to recruit and match suitable IoT devices to crowdsourcing task in dynamic and large IoT networks. These systems include a highly convex formulation with an Integer Linear Program (ILP) constrained problem and a Mixed Integer Linear Program (MILP). Because this problem is NP-complete, several stochastic approaches were also created based on either, Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Optimal Stopping Strategy, and tweaked bipartite graph matching.

IoT Agro Environnemental system for agriculture irrigation using low power wireless sensor network (Industry)

  • Was part of a team that designed a smart IoT application relying on wireless sensor network for data acquisition and control for agriculture irrigation.

Smart Home Security system for home control and monitoring (Undergrad)

  • Designed a Smart Home hybrid mobile application that enables distant control over the house’ facilities (e.g., Windows) and monitor the security status.

Indoor/Outdoor Localization using images (Undergrad)

  • Designed a localization system which uses the images captured by the mobile of the user and gives the latter the ability to determine his/her position and to navigate even in GPS dead zones using image feature matching.