Jean-Sébastien Lerat, a Belgian computer scientist, interested in Cybersecurity and Artificial Intelligence (AI). Armed with a Master's degree in Computer Sciences (2012) from the University of Mons (UMONS), Belgium, John's academic journey saw him venture into additional coursework at University of Brussels (ULB) and University of Louvain-La-Neuve (UCL), further enriching his knowledge base.
His career trajectory has been marked by a deep commitment to education. Starting as a teaching assistant at ULB (2012), he discovered his passion for teaching, eventually leading to his current role as an Associate Professor (2015) at Haute École en Hainaut (HEH). His dedication to education is reinforced by his attainment of the "Certificat d'Aptitude Pédagogique Approprié à l'Enseignement Supérieur" (2017) from UCL, a certification required in Belgium.
His research interests span two primary domains. At UMONS, he delves into the cutting-edge fields of distributed deep learning and Industry 4.0, pioneering innovative technologies with practical applications. In parallel, his work at HEH centers on malware analysis with artificial intelligence, addressing the ever-evolving landscape of digital threats.
Continuing his quest, Jean-Sébastien earned his engineering degree at UMONS (2020) and he is currently pursuing a Ph.D., further elevating his expertise while contributing to the academic community. His research contributions are underscored by a Google H-factor (and I10) of 3, reflecting his significant presence in academic literature.
Looking forward, he remains dedicated to education and research, with future goals encompassing further exploration of distributed deep learning, Industry 4.0, and the critical intersection of AI and cybersecurity. His work continues to shape the technological landscape, ensuring a safer and more advanced digital world.
Distributed Deep Learning and Model Compression for Computer Vision and Industry 4.0 Applications
ABSTRACT :
The rapid advancement of deep learning techniques has revolutionized the landscape of industrial applications, particularly in the domain of Industry 4.0. This thesis addresses the critical challenge of harnessing the power of deep learning for predictive maintenance, defect detection, defect identification, and computer vision in industrial settings. The primary objective is to accelerate the training of deep learning models while ensuring their suitability for resource-constrained edge devices.
The research focuses on the efficient distribution of deep learning tasks, encompassing both training and inference processes, across High-Performance Computing (HPC) and cloud infrastructures. By leveraging the computational capabilities of these infrastructures, this work aims to significantly reduce the time required for model training. Moreover, this approach enables rapid prediction, which is ideal for real-time industrial processes.
In parallel, this thesis investigates the integration of model compression techniques during distributed training. Model compression techniques are crucial for reducing the memory and computational requirements of deep learning models, enabling their deployment on edge devices. These compressed models remain capable of maintaining high predictive accuracy, despite resource limitations, thereby expanding the scope of practical industrial applications.
Through a comprehensive exploration of the proposed approaches, this research contributes valuable insights and practical guidelines for the efficient distribution of deep learning in Industry 4.0 applications. The findings of this thesis not only accelerate the training process of advanced industrial solutions but also pave the way for cost-effective and real-time deployment of deep learning models on edge devices, thereby enhancing the productivity and reliability of industrial systems.
KEYWORDS : Deep learning, High Performance Computing, Industry 4.0, Compression, Compter Vision, Distributed Deep Learning
Cross-biome comparison of microbial association networks
Frontiers in microbiology 6, 1200, 2015
Access publication
Cross-biome comparison of microbial association networks. Front Microbiol 6: 1200
Access publication
Evolution of Common-Pool Resources and Social Welfare in Structured Populations
International Joint Conference on Artificial Intelligence, 2848-2854, 2013
Access publication
Single node deep learning frameworks: Comparative study and CPU/GPU performance analysis
Concurrency and Computation: Practice and Experience 35 (14), e6730, 2023
Access publication
Distributed deep learning: From single-node to multi-node architecture
Electronics 11 (10), 1525, 2022
Access publication
Relational Wide Column Store: a new Big Data paradigm
Proceedings of the 20th ACM International Conference on Computing Frontiers …, 2023
Access publication
