- Presenter: Dr Hamza Ouarnoughi
- Date: Feb 13, 2025
- Abstract: Automated Machine Learning (AutoML) is gaining traction, particularly Neural Architecture Search (NAS), which automates the design of deep learning models. While NAS has advanced image classification and object detection, deploying these complex architectures in resource-constrained environments (e.g., IoT, mobile devices) remains challenging. Hardware-aware NAS (HW-NAS) addresses this by incorporating multi-objective optimization for factors like latency, energy use, and memory, though it introduces new complexities. This presentation will give an overview of our work on AutoML for embedded systems, especially HW-NAS for edge devices.
- Presenter: Dr Kaya Oguz
- Date: Oct 16, 2024
- Abstract: Artificial Intelligence and advanced algorithms play a pivotal role in solving complex problems across various domains. This talk will explore how AI techniques, including machine learning, genetic algorithms, and procedural content generation, can be applied to fields such as speech emotion recognition and procedural content generation in computer games. The session will also cover the APAL algorithm for detecting overlapping communities in biological and social networks. The goal is to present a range of application areas and foster connections with the audience for potential collaborations
"Fuzzy Logic Applications are a powerful tool for handling uncertainty and imprecision in various domains. Explore their diverse use in control systems, artificial intelligence, decision-making, and more. Witness how this flexible approach adapts to real-world scenarios, enabling better, more intuitive solutions in a wide range of applications."
Abstract: "Aircraft and spacecraft are highly complex systems, comprising various subsystems such as flight control, propulsion, and power. These subsystems feature Multiple Input & Multiple Output (MIMO) configurations and exhibit highly nonlinear characteristics. Over recent years, there has been a growing demand for aerospace systems with improved performance, fault tolerance, and enhanced autonomous capabilities. This presentation explores how fault diagnosis, prognosis and artificial intelligence can address these requirements. It delves into model-based, data-driven (artificial intelligence), and hybrid approaches, which have been proposed for fault diagnosis and prognosis. Furthermore, the author has proposed novel methods to meet these evolving demands. These innovative techniques, including the Covariance-based adaptive unscented Kalman filter (CAUKF), Binary grid covariance adaptive Kalman filter (GAUKF), Reinforced Unscented Kalman Filter (an integration of UKF and Reinforcement Learning techniques), and Growing Neural Networks (GNN), hold promise for aerospace system enhancement. The presentation will feature case studies illustrating the application of these methods in spacecraft attitude and orbit control systems, as well as aircraft engines. By examining the latest advancements and methodologies, attendees will gain insights into the pivotal role these techniques play in enhancing aerospace system reliability and efficiency, ultimately addressing current research challenges and shaping the trajectory of future aerospace systems."