Abstract Brain computer interfaces (BCIs) have emerged as a transformative technology enabling direct communication between the human brain and external devices. Central to their effectiveness is the accurate decoding of electroencephalography (EEG) signals, which encapsulate complex neural dynamics... Brain computer interfaces (BCIs) have emerged as a transformative technology enabling direct communication between the human brain and external devices. Central to their effectiveness is the accurate decoding of electroencephalography (EEG) signals, which encapsulate complex neural dynamics across spatial and temporal scales. However, EEG signals are inherently noisy, high-dimensional, non-stationary, and characterized by irregular spatial structures, making their analysis particularly challenging. Traditional deep learning approaches, including convolutional neural networks (CNNs) and recurrent neural networks (RNNs), have demonstrated considerable success in EEG classification tasks. Nevertheless, these models operate under Euclidean assumptions and fail to adequately capture the intrinsic non-Euclidean topology of brain connectivity. In contrast, Graph Neural Networks (GNNs) provide a principled framework for modeling such data by explicitly incorporating relationships between EEG electrodes through graph structures. In this work, EEG signals are modeled as graphs, where electrodes correspond to nodes and functional relationships define edges. A comprehensive formulation of GNN-based learning is presented, including spectral and spatial graph convolutions, attention mechanisms, and spatiotemporal extensions. Furthermore, embedding space analysis is discussed to enhance the interpretability of learned representations. The paper synthesizes recent advancements in graph-based EEG modeling while highlighting key challenges and future research directions. Read More Read Less

Abstract Clustering categorical microbial data presents significant methodological challenges due to the absence of inherent metric structures and the limitations of conventional numerical algorithms. This study systematically evaluates clustering strategies for a categorical dataset of approximately 200 bacterial... Clustering categorical microbial data presents significant methodological challenges due to the absence of inherent metric structures and the limitations of conventional numerical algorithms. This study systematically evaluates clustering strategies for a categorical dataset of approximately 200 bacterial species, characterized by taxonomic, ecological, and pathogenic attributes. We compare a traditional K-Means approach applied to one-hot encoded features with a distance-aware hierarchical clustering framework utilizing Gower dissimilarity. Internal validation metrics, bootstrap-based stability analysis, and multidimensional scaling projections are employed to assess clustering quality, robustness, and biological interpretability. Results indicate that K-Means yields a modest silhouette score (0.19) and moderate stability for a two-cluster solution, reflecting the natural ecological continuity and overlapping niches of microbial taxa rather than algorithmic failure. In contrast, Gower-based hierarchical clustering substantially improves cluster cohesion and separation (silhouette score: 0.34), naturally revealing a robust binary partition primarily governed by human pathogenicity. Density-based methods (DBSCAN) perform poorly due to high sparsity and the absence of well-defined neighborhoods in categorical feature spaces. Visualization and centroid analysis confirm that functional traits and ecological specialization outweigh taxonomic family in driving bacterial similarity. This study demonstrates that aligning distance metrics with categorical data structures is essential for uncovering biologically meaningful patterns. Gower-based hierarchical optimization emerges as a superior framework for microbial classification, offering enhanced stability, interpretability, and alignment with contemporary ecological principles. Read More Read Less

Abstract The rapid integration of artificial intelligence into critical societal domains necessitates robust regulatory frameworks, yet the EU AI Act’s high-risk classifications remain primarily descriptive, lacking quantitative and structural analysis. This study addresses this gap by introducing an... The rapid integration of artificial intelligence into critical societal domains necessitates robust regulatory frameworks, yet the EU AI Act’s high-risk classifications remain primarily descriptive, lacking quantitative and structural analysis. This study addresses this gap by introducing an integrated analytical framework that combines semantic representation with machine learning and multi-dimensional risk modeling. By transforming the eight high-risk AI categories defined in Annex III of the EU AI Act into numerical representations using TF-IDF vectorization, the framework applies K-means clustering and Principal Component Analysis (PCA) to uncover latent structural relationships. The results reveal a low-dimensional semantic space, yielding three coherent clusters: governance and state authority systems, socio-economic decision systems, and technical and safety systems. To capture risk complexity beyond semantic similarity, a composite quantitative risk-scoring model is developed that integrates impact domain, risk type, decision criticality, and degree of human impact. This multi-dimensional approach demonstrates that semantic proximity does not equate to equivalent risk severity, with governance systems exhibiting the highest composite risk scores. Rigorous validation through silhouette analysis, inter-cluster separation metrics, and stability testing confirms the framework’s reliability. By bridging regulatory classification with quantitative risk evaluation, this study provides a scalable, interpretable tool for policymakers and practitioners. The findings advocate for differentiated, data-driven regulatory strategies that align oversight mechanisms with the distinct structural and risk profiles of high-risk AI systems. Read More Read Less