Yassine Hmidy

I teach a Master’s level course in Information Systems and Decision, where I explore how technology and data support strategic and operational decision-making. With a background in artificial intelligence and applied mathematics, my work bridges theory and practice, helping students and professionals understand the critical role of information systems in organizational success.

As the Research Project Manager for the AI powered Big data for Banking project, I led the scientific efforts to develop AI-driven solutions for fraud detection. This high-impact initiative aimed to enhance fraud prevention strategies by leveraging cutting-edge machine learning models and data analytics. I was responsible for managing a cross-disciplinary team of 25 consultants, ensuring that each team member was tasked with work aligned with their expertise and the project's objectives. My role involved overseeing the scientific direction of the project, guiding the research methodology, and ensuring the integration of AI models into the business context. In addition to managing the team, I regularly presented project updates, milestones, and results to key stakeholders, effectively communicating the progress and potential business impact of the AI models we developed. This experience not only strengthened my leadership and project management skills but also deepened my expertise in applying AI to complex, real-world problems such as fraud detection.

The aim of my thesis is to build a new architecture of neural networks that would avoid overfitting and save computation cost. This new neural network has no activation function needed. The non linearity is obtained throug the composition of fuzzy aggregation functions.

I have taught first-year computer science courses, including: UNIX: Introducing students to operating system concepts, shell scripting, and system commands. Data Structures and Algorithms: Covering foundational topics like linked lists, trees, and algorithmic design principles. Advanced Graph Theory and Combinatorial Optimization: Exploring complex optimization problems and advanced graph algorithms.

As a Machine Learning Engineer at Thales Alenia Space, I worked on improving the accuracy of Feature Trust Measurement (FTM) for hyperspectral images. My primary task involved designing and implementing Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) to analyze large hyperspectral datasets. The objective was to identify why certain images were yielding incorrect FTM estimations and to enhance the overall reliability of these predictions. To address this, I also applied unsupervised deep learning techniques, specifically Self-Organizing Maps (SOMs), to cluster the data and uncover underlying patterns that were influencing erroneous estimations. This approach provided a deeper understanding of the dataset and improved the trustworthiness of the model's estimations. The project sharpened my skills in deep learning, computer vision, and unsupervised learning, while contributing to Thales Alenia Space's efforts to advance satellite image processing technologies.