http://hdl.handle.net/2268.2/11098 Thu, 05 Mar 2026 22:47:08 GMT 2026-03-05T22:47:08Z http://hdl.handle.net/2268.2/25196 Title: Research-Thesis: Disaster Response Improvement Using Heuristic Methods Abstract: Disaster response operations require fast and reliable planning under uncertainty and time pressure. This thesis addresses this challenge by studying and developing a robust time-sensitive capacitated orienteering problem (RT-CTOP) tailored to humanitarian logistics. The objective is to maximize collected rewards representing disaster victims, while accounting for uncertain travel times, limited resources, and time-dependent service values. A robust optimization framework based on a budgeted uncertainty set is proposed to explicitly model travel-time variability. As is well known, exact methods such as Mixed-Integer Linear Programming (MILP) are NP-hard for this class of problems, meaning that they quickly become computationally intractable as instance size increases. To overcome this limitation, two metaheuristic approaches, Genetic Algorithm (GA) and Variable Neighborhood Search (VNS), are designed and adapted to the RT-CTOP formulation. The results show that GA and VNS consistently deliver high-quality and stable solutions while significantly outperforming the MILP in terms of scalability and computational behavior. In instances where the MILP is solvable, heuristic solutions closely approximate optimal values. Under tight runtime limits or increased uncertainty, heuristic methods maintain solution quality, whereas MILP performance degrades sharply. The findings confirm the effectiveness of metaheuristic approaches in addressing large-scale and time-critical disaster response routing problems under uncertainty, and establish a foundation for future research on adaptive and data-driven robust optimization frameworks. Tue, 13 Jan 2026 23:00:00 GMT http://hdl.handle.net/2268.2/25196 2026-01-13T23:00:00Z http://hdl.handle.net/2268.2/22785 Title: LLM Size Reduction & Carbon Footprint Abstract: In recent years, model compression techniques have proven highly effective at reducing large language models’ storage footprint and accelerating inference. By lowering memory requirements and increasing computational speed, these methods offer a pathway to more energy-efficient large models. However, altering model parameters through compression inherently risks degrading performance. Thus, the trade-off between efficiency gains and potential quality loss remains central when adopting compression strategies. While the environmental impact of training large models has received growing attention, the effects of compression on hardware energy consumption and related GHG emissions during inference remain largely unexplored. To address this gap, I conducted an empirical study investigating the effects of compression on both model performance and energy consumption across hardware components—specifically CPU, GPU, and RAM. Four decoder-only transformer models were selected for their academic relevance and maturity: LLaMA-7B, LLaMA-30B, Mistral-7B-v0.3, and Mistral Small 3. Each model was compressed using the OPTQ method at 4-bit precision. Model performance was evaluated using WikiText-2 perplexity, and MMLU and IFEval accuracy. Energy consumption was measured using CodeCarbon during inference on two high-end hardware configurations. To contextualize these results, I conducted a comparative carbon footprint analysis using four electricity mixes, offering a grid-aware perspective on compression-related emissions in CO₂eq. The findings show that (1) quantization-induced performance degradation is marginal; compressed models retain nearly all capabilities across benchmarks. (2) Compression can substantially reduce energy use, but the magnitude depends on hardware—one configuration yielded up to 39% savings, while another saw a 26% increase. (3) Environmental impact hinges not only on model and energy use but also on deployment geography: a compressed model on a carbon-intensive grid can emit up to six times more than a full-sized model on a clean grid. Thus, sustainability benefits of compression must be assessed in relation to hardware and geographic context. Thu, 19 Jun 2025 22:00:00 GMT http://hdl.handle.net/2268.2/22785 2025-06-19T22:00:00Z http://hdl.handle.net/2268.2/21578 Title: Development and Evaluation of a 3D Human Body Visualization Tool Using HoloLens 2 for Medical Education at the University of Liège Abstract: This thesis presents the development and evaluation of a 3D human body visualization tool utilizing HoloLens 2 technology for medical education at the University of Liège. The primary objective was to assess the effectiveness of this mixed reality (MR) technology in enhancing the visualization, comprehension, and manipulation of anatomical structures, specifically in the context of medical education. The research began with the adaptation of a pre-existing augmented reality (AR) application, originally designed for smartphones, which allowed users to explore a 3D model of the human body. The application was modified to function on the HoloLens 2 platform, emphasizing interactive and immersive learning experiences A qualitative research methodology was employed to gather insights from medical students at various stages of their education. The study was divided into two phases: the first phase focused on evaluating the general experience with the HoloLens 2 application, and the second phase involved a comparison with the existing mobile AR anatomy application. Students provided feedback on usability, engagement, and the educational value of the tool. The findings showed that HoloLens 2 greatly improved students' ability to visualize and understand spatial relationships in anatomy, especially in complex areas like vascularization and innervation. The intuitive interface enabled quick adaptation, despite initial calibration and gesture challenges. While the immersive experience boosted engagement, there was some concern that the technology might be seen more as a novelty than a serious educational tool. The feasibility of implementing the HoloLens 2 as a regular tool in medical education was also analyzed through a SWOT analysis. The strengths of the technology include its intuitive user interface, the ability to visualize complex anatomical relationships in 3D, and the enhanced engagement it provides to students. The weaknesses identified include limitations in user experience, hardware, resources, and environment, and the financial cost associated with the technology. Opportunities for the HoloLens 2 include its potential to revolutionize medical education by providing a more interactive and immersive learning environment, and the possibility of its use in remote learning scenarios. However, there are also threats, such as direct and indirect competitors and the uncertain future of the HoloLens technology itself. In conclusion, the HoloLens 2 shows great promise as a transformative tool for medical education, offering enhanced visualization and engagement through its advanced MR capabilities. However, to fully realize its potential, further refinement of the technology and careful consideration of its integration into existing curricula are necessary. The SWOT analysis suggests that while there are challenges to adoption, the opportunities for improving medical education are significant. Sun, 01 Sep 2024 22:00:00 GMT http://hdl.handle.net/2268.2/21578 2024-09-01T22:00:00Z http://hdl.handle.net/2268.2/20248 Title: Comparaison de 4 LLM dans le contexte du traitement du langage naturel Abstract: À l'ère de l'intelligence artificielle, les modèles de langage de grande taille (LLM) jouent un rôle crucial dans le traitement du langage naturel (NLP). Ce mémoire compare quatre LLM notables GPT, Mistral, Falcon et Llama, sur différents critères pertinents : la taille, l'architecture des modèles, le type de licence utilisée, les langues supportées par les modèles, l'èthique, la notoriété, la performance, la tendance à halluciner et les bases de donées utilisées par les modèles pour leur entrainement. Mon, 17 Jun 2024 22:00:00 GMT http://hdl.handle.net/2268.2/20248 2024-06-17T22:00:00Z