http://hdl.handle.net/2268.2/6046 Tue, 21 Apr 2026 19:31:19 GMT 2026-04-21T19:31:19Z http://hdl.handle.net/2268.2/24931 Title: Biomedical Text Classification Using LSTM, GRU and Bahdanau Attention Abstract: This research evaluates the performance of deep learning models—Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM), with Bahdanau attention added—for biomedical text classification using two datasets: the Memorial Sloan Kettering Cancer Center (MSKCC ) dataset for binary classification of genetic mutations and the Medical Abstracts dataset for multi-class disease categorization. The experiments showed that GRU models generally offer better training efficiency and balanced accuracy than LSTM models, and that class weighting proved more effective than Synthetic Minority Over-sampling Technique (technique for oversampling minority classes) (SMOTE) in handling class imbalance. While few-shot learning remains challenging, models using Biomedical Natural Language Processing (BioMedNLP) contextual embeddings combined with attention mechanisms demonstrated promising generalization, particularly in low-resource scenarios. These findings support the development of more robust and equitable Natural Language Processing (NLP) systems for biomedical applications. Sun, 07 Sep 2025 22:00:00 GMT http://hdl.handle.net/2268.2/24931 2025-09-07T22:00:00Z http://hdl.handle.net/2268.2/24781 Title: Can Large Language Models accelerate the correction of student code ? Abstract: This thesis assesses to what extent large language models (LLMs) can accelerate the correction of student code in an introductory C programming course. The motivation is practical. Autograders are helpful grading tools but they miss many dimensions of code quality like clarity, efficiency and style. Hence, human review remains heavy and slow. LLMs, which can read code in context and provide natural language feedback, may fill part of this gap. The principal objective is to determine how they can be leveraged to accelerate code correction. We conduct three sets of experiments on real coursework from the ``Additional Information Theory'' course at the University of Liège. First, we run preliminary code-generation tests to determine whether state-of-the-art LLMs can solve the course tasks. Second, we evaluate automated grading with Qwen2.5-Coder-7B on two datasets. These sets respectively consist of student submissions for a homework assignment and a project. We compare model-predicted grades and feedback to human grades. Third, we study error detection and code correction on the same homework by fine-tuning Qwen2.5-Coder-7B with LoRA using prompt-response pairs. With respect to grading, the model's numeric predictions are not reliable. On both tasks, the mean errors often match or exceed those obtained by a constant baseline. However, when the task is reframed as a simpler classification problem where we ask the LLM whether each submission is fully correct, Qwen's performance is above chance. The best setting is the one in which we use a criteria-based prompt in French. This method consistently outperforms the baseline. Nevertheless, it remains insufficient for autonomous grading. In error detection and correction, our initial fine-tuning with Qwen-generated data slightly improved correction rates. However, it often produced full code rewrites rather than genuine code corrections. A second fine-tuning attempt used more diverse, high-quality training data generated by OpenAI models, which encouraged targeted edits. However, this reduced correction performance on student submissions. These results indicate that improving a model's error detection and repair abilities is difficult with such limited datasets. Overall, we find that LLMs are not powerful enough yet to replace human graders for either grading or error detection and correction. Their most promising use today is as a support tool alongside autograders and human review. Still, our findings are bounded by scope as we only used tasks in C from a specific course and minimal prompting. We recommend exploring more powerful models and considering fine-tuning on Python tasks with a larger, more comprehensive training set. Sun, 07 Sep 2025 22:00:00 GMT http://hdl.handle.net/2268.2/24781 2025-09-07T22:00:00Z http://hdl.handle.net/2268.2/22458 Title: Injury risk factors in volleyball Abstract: Based on several kind of performed tests as well as practical and physical information on a sample of a few hundreds volleyball players analysed and followed-up between 2018 and 2022, determine the risk of injury of a lambda individual and, more generally, what are the injury risk factors in volleyball. Thu, 23 Jan 2025 23:00:00 GMT http://hdl.handle.net/2268.2/22458 2025-01-23T23:00:00Z http://hdl.handle.net/2268.2/21150 Title: Evaluating LLMs on large contexts : a RAG approach on text comprehension Abstract: While the latest Large Language Models (LLMs) continue to expand in size and context window capacity, their knowledge base remains constrained by their training corpus. Retrieval Augmented Generation (RAG) offers a solution to this limitation by enhancing the LLM’s responses with relevant information retrieved from external sources. In contrast to the rapidly growing context windows, which now extend to millions of tokens, this study evaluates the effectiveness of augmenting prompts as an alternative approach to using large contexts, this is done by evaluating multiple-choice questions originally made for long context settings. By using parts of the context with RAG, I demonstrate that a well-constructed RAG system can achieve strong performance with significantly reduced token usage. However, the results also reveal challenges related to prompt sensitivity. Despite these challenges, the potential reduction in inference costs due to lower token usage makes this approach particularly appealing, depending on the application context. Wed, 04 Sep 2024 22:00:00 GMT http://hdl.handle.net/2268.2/21150 2024-09-04T22:00:00Z