Doctorate thesis defense on May 23^th 2026 at 10H00 AM ,in Amphitheater Ibn Khaldoun, SUP'COM 2.

Entitled: Embedded Lightweight Artificial Intelligence for IoT Systems: Optimization Methods and Application

Presented by : Yassmine BEN DHIAB

Thesis Committee

Abstract

This thesis addresses the efficient deployment of Artificial Intelligence in Internet of Things environments where devices operate under strict constraints in terms of computing power, memory, energy, and connectivity. Rather than relying on centralized cloud infrastructures, this work investigates how AI can be executed closer to the data source, directly at the network edge, enabling faster, more private, and more resource-aware processing. The research adopts a holistic optimization perspective that spans the entire AI pipeline, from data acquisition and preprocessing to model architecture design and system-level deployment decisions. The overarching goal is to make Edge AI more practical, lightweight, and responsive, while remaining suitable for real-world, resource-constrained applications. The contributions are structured along three complementary axes: improving data representation and dataset quality to enhance model robustness under constrained conditions; designing and benchmarking lightweight AI models for diverse applications including human activity recognition, epileptic seizure monitoring, and plant disease classification; and developing adaptive security mechanisms tailored to healthcare IoT systems. Collectively, this thesis demonstrates that successful Edge AI demands more than high predictive accuracy. It requires a careful and principled co-optimization of performance, computational efficiency, real-time responsiveness, privacy, and security. By jointly addressing data, model, and system dimensions, this work advances the development of reliable, deployable AI solutions for the next generation of constrained IoT environments.

 Keywords: Edge AI; Internet of Things; Embedded Artificial Intelligence; TinyML; Resource-Constrained Devices; Lightweight Deep Learning; Data-, Model-, and System-Centric Optimization; Adaptive Security.