Search results

Synopsis deep learning methods have become increasingly popular for building intelligent systems. Currently, many deep learning architectures constitute the state of the art in their respective domains, such as image recognition, text generation, speech recognition, etc. The availability of mature libraries and frameworks to develop such systems is also a key factor in this success.This work explores the use of these architectures to build intelligent systems for sign languages. The creation of large sign language data corpora has made it possible to train deep learning architectures from scratch. The contributions presented in this work cover all aspects of the development of an intelligent system based on deep learning. A first contribution is the creation of a database for the Langue des Signes de Belgique Francophone (LSFB). This is derived from an existing corpus and has been adapted to the needs of deep learning methods. The possibility of using crowdsourcing methods to collect more data is also explored.The second contribution is the development or adaptation of architectures for automatic sign language recognition. The use of contrastive methods to learn better representations is explored, and the transferability of these representations to other sign languages is assessed.Finally, the last contribution is the integration of models into software for the general public. This led to a reflection on the challenges of integrating an intelligent module into the software development life cycle.Jury membersProf. Wim VANHOOF, President, University of NamurProf. Benoît FRENAY, Promoter, University of NamurProf. Anthony CLEVE, Co-promoter, University of NamurProf. Laurence MEURANT, Internal Member, University of NamurProf. Lorenzo BARALDI, External Member, University of ModenaProf. Annelies BRAFFORT, External Member, University of Paris-SaclayProf. Joni DAMBRE, External Member, University of GhentYou are cordially invited to a drink, which will follow the public defense. For a good organization, please give your answer by Friday June 6.

Space has become a major economic and strategic issue. As a member of the European UNIVERSEH Alliance, UNamur explores this space theme in its various departments, from physics to geology, via mathematics, computer science or philosophy. Without forgetting to address the general public, who still dream of the stars...

Can a computer learn a language like a child? A recent study published in the leading journal Computational Linguistics by Professors Katrien Beuls (Université de Namur) and Paul Van Eecke (AI-lab, Vrije Universiteit Brussel) sheds new light on this question. The researchers argue for a fundamental revision of the way artificial intelligence acquires and processes language.

For Bruno Dumas, computer science fits in with the principles of applied psychology Artificial intelligence (AI) is interfering in our professional as well as our private lives. It both seduces and worries us. On a global scale, it is at the heart of major strategic, societal or economic issues, still being debated in mid-February 2025, at the AI World Summit in Paris. But how can we, as users, avoid being subjected to it? How can we gain access to the necessary transparency of its workings? By placing his research prism on the user's side, Bruno Dumas is something of a "computer psychologist". An expert in human-computer interaction, co-president of the NaDI Institute (Namur Digital Institute), he defends the idea of a reasoned and enlightened use of emerging technologies.

Abstract Deep learning has become an extremely important technology in numerous domains such as computer vision, natural language processing, and autonomous systems. As neural networks grow in size and complexity to meet the demands of these applications, the cost of designing and training efficient models continues to rise in computation and energy consumption. Neural Architecture Search (NAS) has emerged as a promising solution to automate the design of performant neural networks. However, conventional NAS methods often require evaluating thousands of architectures, making them extremely resource-intensive and environmentally costly.This thesis introduces a novel, energy-aware NAS pipeline that operates at the intersection of Software Engineering and Machine Learning. We present CNNGen, a domain-specific generator for convolutional architectures, combined with performance and energy predictors to drastically reduce the number of architectures that need full training. These predictors are integrated into a multi-objective genetic algorithm (NSGA-II), enabling an efficient search for architectures that balance accuracy and energy consumption.Our approach explores a variety of prediction strategies, including sequence-based models, image-based representations, and deep metric learning, to estimate model quality from partial or symbolic representations. We validate our framework across three benchmark datasets, CIFAR-10, CIFAR-100, and Fashion-MNIST, demonstrating that it can produce results comparable to state-of-the-art architectures with significantly lower computational cost. By reducing the environmental footprint of NAS while maintaining high performance, this work contributes to the growing field of Green AI and highlights the value of predictive modelling in scalable and sustainable deep learning workflows. Jury Prof. Wim Vanhoof - University of Namur, BelgiumProf. Gilles Perrouin - University of Namur, BelgiumProf. Benoit Frénay - University of Namur, BelgiumProf. Pierre-Yves Schobbens - University of Namur, BelgiumProf. Clément Quinton - University of Lille, FranceProf. Paul Temple- University of Rennes, FranceProf. Schin'ichi Satoh - National Institute of Informatics, Japan

Abstract Deep learning has become an extremely important technology in numerous domains such as computer vision, natural language processing, and autonomous systems. As neural networks grow in size and complexity to meet the demands of these applications, the cost of designing and training efficient models continues to rise in computation and energy consumption. Neural Architecture Search (NAS) has emerged as a promising solution to automate the design of performant neural networks. However, conventional NAS methods often require evaluating thousands of architectures, making them extremely resource-intensive and environmentally costly.This thesis introduces a novel, energy-aware NAS pipeline that operates at the intersection of Software Engineering and Machine Learning. We present CNNGen, a domain-specific generator for convolutional architectures, combined with performance and energy predictors to drastically reduce the number of architectures that need full training. These predictors are integrated into a multi-objective genetic algorithm (NSGA-II), enabling an efficient search for architectures that balance accuracy and energy consumption.Our approach explores a variety of prediction strategies, including sequence-based models, image-based representations, and deep metric learning, to estimate model quality from partial or symbolic representations. We validate our framework across three benchmark datasets, CIFAR-10, CIFAR-100, and Fashion-MNIST, demonstrating that it can produce results comparable to state-of-the-art architectures with significantly lower computational cost. By reducing the environmental footprint of NAS while maintaining high performance, this work contributes to the growing field of Green AI and highlights the value of predictive modelling in scalable and sustainable deep learning workflows. Jury Prof. Wim Vanhoof - University of Namur, BelgiumProf. Gilles Perrouin - University of Namur, BelgiumProf. Benoit Frénay - University of Namur, BelgiumProf. Pierre-Yves Schobbens - University of Namur, BelgiumProf. Clément Quinton - University of Lille, FranceProf. Paul Temple- University of Rennes, FranceProf. Schin'ichi Satoh - National Institute of Informatics, Japan

Abstract Deep learning has become an extremely important technology in numerous domains such as computer vision, natural language processing, and autonomous systems. As neural networks grow in size and complexity to meet the demands of these applications, the cost of designing and training efficient models continues to rise in computation and energy consumption. Neural Architecture Search (NAS) has emerged as a promising solution to automate the design of performant neural networks. However, conventional NAS methods often require evaluating thousands of architectures, making them extremely resource-intensive and environmentally costly.This thesis introduces a novel, energy-aware NAS pipeline that operates at the intersection of Software Engineering and Machine Learning. We present CNNGen, a domain-specific generator for convolutional architectures, combined with performance and energy predictors to drastically reduce the number of architectures that need full training. These predictors are integrated into a multi-objective genetic algorithm (NSGA-II), enabling an efficient search for architectures that balance accuracy and energy consumption.Our approach explores a variety of prediction strategies, including sequence-based models, image-based representations, and deep metric learning, to estimate model quality from partial or symbolic representations. We validate our framework across three benchmark datasets, CIFAR-10, CIFAR-100, and Fashion-MNIST, demonstrating that it can produce results comparable to state-of-the-art architectures with significantly lower computational cost. By reducing the environmental footprint of NAS while maintaining high performance, this work contributes to the growing field of Green AI and highlights the value of predictive modelling in scalable and sustainable deep learning workflows. Jury Prof. Wim Vanhoof - University of Namur, BelgiumProf. Gilles Perrouin - University of Namur, BelgiumProf. Benoit Frénay - University of Namur, BelgiumProf. Pierre-Yves Schobbens - University of Namur, BelgiumProf. Clément Quinton - University of Lille, FranceProf. Paul Temple- University of Rennes, FranceProf. Schin'ichi Satoh - National Institute of Informatics, Japan

Japan is almost 10,000 kilometers from Belgium, a country that fascinates, not least for its rich culture full of contrasts. Researchers at UNamur maintain close ties with several Japanese institutions, particularly in the fields of computer science, mathematics and video games. Let's take a look at some of these collaborations..