https://www.um.edu.mt/library/oar/handle/123456789/134219 2026-06-03T11:54:30Z 2026-06-03T11:54:30Z https://www.um.edu.mt/library/oar/handle/123456789/146931 2026-05-29T08:00:32Z 2025-01-01T00:00:00Z

Title: Leveraging invariant prediction for mitigating specificity constraints in affect modelling Abstract: Affect modelling aims to predict human emotional states from multimodal signals, yet current approaches often struggle to generalise beyond the specific datasets or contexts in which they are trained. This dissertation investigates the use of in variant features, predictors whose relationship with affective states remains stable across distinct environments, as a strategy to improve generalisability. To this end, two publicly available corpora, AGAIN and RECOLA, were systematically parti tioned into environments defined by user, task, and annotator triplets. An envi ronment refers to the conditions under which data is collected, and data gathered within the same environment is assumed to come from the same underlying distri bution. The Invariant Causal Prediction (ICP) framework was employed to identify stable features across these environments, which were then compared against full feature sets and principal components derived through PCA. Three supervised learning models—Logistic Regression, a feed-forward Neural Network,and a Long Short-Term Memory (LSTM)network — were trained under all three feature conditions, using group-based cross-validation to avoid information leakage. Results demonstrate that invariant features can deliver measurable benefits for feed-forward models, particularly in enhancing accuracy and correlation while substantially reducing feature dimensionality. However, their advantages were less consistent for sequence models like LSTMs, where temporal dependencies were not fully captured by invariants alone. Statistical significance tests further showed that invariant features improved balanced classification (F1) more strongly in AGAIN than in RECOLA,underscoring the dataset-specific nature of their effectiveness. Overall, the findings highlight both the promise and the limitations of invariance in affect modelling. While not a universal solution, invariant features represent a principled means of isolating robust predictors across heterogeneous contexts, contributing to the broader goal of developing affective systems that are reliable, interpretable, and adaptable across diverse real-world settings. Description: M.Sc.(Melit.)

2025-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/146930 2026-05-29T07:59:03Z 2025-01-01T00:00:00Z

Title: Siamese network‐based vector embeddings of MRI scans for twin identification Abstract: Monozygotic twins are identical twins that develop from a single fertilised egg that spontaneously splits, resulting in two individuals sharing 100% genetic material. Identifying monozygotic twins from brain MRI scans represents a frontier challenge in computational medical imaging with significant implications for understanding genetic influences on neuroanatomical structure through direct pattern recognition. While classical twin studies using ACE models decompose statistical variance to establish independent regional heritability estimates (60‐80%), this study introduces a fundamentally different computational framework that learns directly from MRI data to rank neuroanatomical regions by their collective discriminative capacity for genetic similarity detection, complementing traditional statistical approaches through data‐driven analysis. Adeep learning methodology employing Siamese networks with 3D CNN backbones is developed for automated twin identification using 138 genetically verified monozygotic twin pairs (276 subjects) from the Human Connectome Project S1200 dataset. Modified U‐Net, ResNet, and DenseNet architectures generate 128‐dimensional embeddings optimised via triplet loss with hard negative mining, forcing models to learn subtle genetic signatures by focusing on challenging discriminative examples that distinguish twins from their most similar morphological matches. U‐Net achieved superior computational performance with 92.0% F1‐score (σ = 2.5%), 95.2% AUC‐ROC, and 91.4% accuracy, while ResNet demonstrated competitive results (89.6% F1‐score) and DenseNet showed greater variability (88.5% F1‐score). Embedding analysis reveals clear bimodal separation between genetically related and unrelated individuals through learned morphological patterns. Layer‐Wise Relevance Propagation analysis provides the first data‐driven ranking of neuroanatomical regions by discriminative importance for genetic relatedness detection. Statistical analysis reveals pronounced subcortical dominance with large effect size (Cohen’s d = 2.80, p = 3.89e‐6), with six subcortical structures occupying top positions, including the thalamus (0.955), brainstem (0.875), and hypothalamus (0.707). This computational hierarchy contrasts with traditional ACE studies reporting highest heritability in cortical areas (frontal 78‐95%, temporal 77‐89%), demonstrating that direct pattern recognition from MRI data identifies different neuroanatomical signatures than statistical variance decomposition. Notably, models utilise practically all brain regions (most importance scores > 0.2), indicating distributed multivariate processing rather than selective regional dependence. Ablation studies confirm data augmentation’s critical role, with substantial i performance improvements across CNN architectures. Clinical integration through standard neuroimaging formats in Connectome Workbench demonstrates immediate practical utility, positioning this computational approach for adoption in research and clinical environments requiring direct analysis of genetic influences in brain structure. The framework advances precision neuroimaging by providing automated, quantitative genetic similarity detection through direct pattern recognition, revealing spatial insights that complement traditional heritability studies while offering methodological advances applicable to diverse medical imaging classification tasks requiring regional discriminative analysis Description: M.Sc.(Melit.)

2025-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/141991 2025-12-05T10:17:45Z 2025-01-01T00:00:00Z

Title: AI‐Driven gesture recognition with smart gloves Abstract: This research presents the development of an AI‐driven gesture recognition system aimed to enhance Human‐Computer Interaction through the use of smart gloves. Many emerging applications, such as virtual reality, robotics, and assistive technologies, require detailed motion capture of the hand in three dimensions. Traditional input devices are not designed to capture such motion, whereas wearable solutions like smart gloves offer a practical means of collecting complex motion data for gesture interpretation. This study proposes a system capable of interpreting dynamic hand gestures captured using smart gloves. A custom dataset was collected using Rokoko smart gloves, recording 14 gesture classes from 14 subjects. Time‐series data captured from the smart gloves was preprocessed, and a range of feature extraction methods, including statistical, frequency‐domain, and motion‐based techniques, were applied. Experimental results were carried out to determine which features or combination of features gives the best result. Dimensionality reduction methods, namely Principal Component Analysis and Autoencoders, were examined to optimise the feature space and reduce complexity. A number of classification models were implemented and compared, including Support Vector Machines, K‐Nearest Neighbours, Hidden Markov Models, as well as, deep learning approaches such as CNN‐LSTM networks. Experimental results showed that while most models achieved high accuracy on validation data, up to 93.64%, performance significantly decreased when tested on data from unseen subjects, dropping to 20.39‐28.93%. This highlights the challenge of inter‐subject generalisation. To mitigate this, personalised models were implemented, showing good performance improvements. The SVM classifiers achieved accuracy results ranging from 67.9% to 92.9%, and the majority of precision, recall, and F1 scores exceeding 85%, while CNN‐LSTM models achieved an accuracy above 95% consistently. Precision, recall, and F1‐score values also remained high. This work contributes to the field of gesture recognition by systematically evaluating feature engineering and modelling techniques on multichannel time‐series data. It underscores the importance of personalised learning strategies and provides insight into the practical limitations of real‐world deployment, such as latency and subject variability. Future work may explore domain adaptation, multimodal sensing, and real‐time implementation to further advance robust gesture‐based interfaces. Description: M.Sc.(Melit.)

2025-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/141983 2025-12-05T09:53:57Z 2025-01-01T00:00:00Z

Title: Table selection using information retrieval techniques for table-agnostic Text-to-SQL Abstract: Text-to-SQL has been effectively addressed using various NLP approaches, enabling the translation of natural language queries into SQL queries. A common prerequisite for these implementations, however, is the availability of the database table during inference. This requirement can pose challenges in scenarios where the table is not readily accessible to users. This work is motivated by the ongoing development of a chatbot tool within a private company, aimed at streamlining database interactions for the users. To address the table accessibility limitation, this study leverages Information Retrieval techniques to implement table selection based solely on the natural language query. We finetune pre-trained models like BERT and GIST-NoInstruct using the ColBERT method. We train our models using data we curate in-house by employing established LLM-prompting techniques. We prepare individual training datasets using two negative sampling techniques: uniform distribution and weighted probability distribution. We also experiment with various data fusion techniques such as RRF, CombMNZ, and Linear Combination to combine results from multiple search strategies. Our approach outperforms baseline methods in table retrieval, while also providing a comparative analysis of various retrieval strategies. Description: M.Sc.(Melit.)

2025-01-01T00:00:00Z