https://www.tmachineintelligence.ir/ Transactions on Machine Intelligence en daily 1 Mon, 02 Mar 2026 00:00:00 +0330 Mon, 02 Mar 2026 00:00:00 +0330 https://www.tmachineintelligence.ir/article_244942.html Virtual machine (VM) allocation remains a foundational challenge in Infrastructure-as-a-Service (IaaS) cloud computing, particularly as multi-tenant environments grow in scale, heterogeneity, and complexity. Recent advancements increasingly emphasize the dual objectives of fairness and optimization, reflecting the need for equitable resource distribution without compromising system performance. This systematic review synthesizes peer-reviewed research published between 2020 and 2025, evaluating fairness-driven, optimization-driven, and hybrid VM allocation models. Following a PRISMA-aligned search strategy across IEEE Xplore, ACM Digital Library, ScienceDirect, SpringerLink, and Google Scholar, 15 high-quality studies meeting strict inclusion criteria were selected. Findings reveal that while fairness-driven models successfully reduce resource starvation and improve equity, they often incur higher makespan and computational overhead. Conversely, optimization-driven models demonstrate substantial gains in throughput, energy efficiency, and load balancing, yet frequently deprioritize fairness, leading to disproportionate resource allocation for low-priority tasks. Hybrid models have subsequently emerged as a dominant research direction, leveraging multi-objective and machine learning-based approaches to balance efficiency and equity effectively. Despite this progress, key research gaps persist, including a lack of standardized fairness metrics, limited validation using real-world workload traces, and inconsistent reporting of scalability and convergence properties. This review situates the Optimized Proportional Equity Model (OPEM) within this evolving landscape, highlighting its deterministic, fairness-centered design and potential to contribute to transparent resource allocation frameworks. The study concludes with actionable recommendations for developing next-generation allocation mechanisms grounded in rigorous evaluation, hybrid optimization, and standardized assessment. https://www.tmachineintelligence.ir/article_244943.html The advent of large language models (LLMs) has revolutionized the integration of knowledge graphs (KGs) in the biomedical and cognitive sciences, effectively overcoming the limitations of traditional machine learning methods in capturing intricate semantic links among genes, diseases, and cognitive processes. This paper introduces MultiCNKG, an innovative framework that merges three distinct knowledge sources: the Cognitive Neuroscience Knowledge Graph (CNKG), containing 2.9K nodes and 4.3K edges across 9 node types and 20 edge types; the Gene Ontology (GO), featuring 43K nodes and 75K edges across 3 node types and 4 edge types; and the Disease Ontology (DO), comprising 11.2K nodes and 8.8K edges with 1 node type and 2 edge types. Utilizing advanced LLMs such as GPT-4, we perform automated entity alignment, semantic similarity computation, and graph augmentation to construct a unified, cohesive KG that interconnects genetic mechanisms, neurological disorders, and cognitive functions. The resulting MultiCNKG unified graph encompasses 6.9K nodes across 5 distinct types and 11.3K edges spanning 7 relational types, establishing a multi-layered analytical pipeline from molecular to behavioral domains. Empirical evaluations demonstrate robust framework performance, achieving an 85.20% precision rate, 87.30% recall, 92.18% coverage, 82.50% graph consistency, a 40.28% novelty detection rate, and an 89.50% expert validation score. Furthermore, link prediction benchmarks utilizing TransE (MR: 391, MRR: 0.411) and RotatE (MR: 263, MRR: 0.395) yield highly competitive performance against standard benchmarks like FB15k-237 and WN18RR. Ultimately, this integrated KG advances clinical and research applications in personalized medicine, cognitive disorder diagnostics, and data-driven hypothesis formulation within cognitive neuroscience. https://www.tmachineintelligence.ir/article_244944.html Effectively fusing heterogeneous semantic information while ensuring training stability remains a critical challenge in graph learning. This paper proposes HGNN-SF, a robust architecture designed for link prediction in dense and complex heterogeneous graphs. The core of the proposed model is a novel Semantic Fusion Layer (SF-Layer) that integrates a two-level attention mechanism. First, Intra-Type Spectral Attention incorporates spectral normalization into relation-specific attention modules to stabilize message propagation and prevent gradient explosion in high-degree, dense graph structures. Second, a Hierarchical Semantic Fusion mechanism dynamically assigns importance weights to aggregated messages from distinct relation types, enabling effective semantic integration without relying on predefined meta-paths. To regulate information flow and mitigate over-smoothing in deep heterogeneous message passing, node representations are updated using a Gated Recurrent Unit (GRU). An ablation study demonstrates that this gated update significantly outperforms conventional residual connections in preserving feature discriminability. Additionally, Hard Negative Sampling is employed during training to enhance model robustness against structurally challenging negative instances. Extensive experiments conducted on four large-scale, real-world datasets (DBLP, ACM, Amazon, and LiveJournal) demonstrate that HGNN-SF consistently outperforms strong graph neural network (GNN) and heterogeneous GNN (HGNN) baselines in link prediction tasks. Notably, on the highly challenging LiveJournal dataset, HGNN-SF achieves a Test AUC of 0.9044, confirming the efficacy of combining spectral attention and gated semantic fusion for stable, accurate representation learning in heterogeneous networks. https://www.tmachineintelligence.ir/article_244946.html Earthquakes pose a catastrophic threat to human life and critical infrastructure, necessitating the development of highly accurate early warning systems. The primary objective of this research is to design an earthquake early warning framework capable of predicting both seismic magnitude and occurrence timing by integrating machine learning and deep learning architectures. In this study, a Random Forest Regression model is employed to estimate earthquake magnitudes, while a time-series Transformer algorithm is utilized to forecast occurrence times. The underlying dataset, sourced from seismic networks in Japan, underwent comprehensive preprocessing, normalization, and train-test partitioning. To predict timing, data was converted into a sequential time-series format before feeding it into the Transformer network. The designed Transformer model incorporates multi-head self-attention mechanisms, layer normalization, and aggregation modules, significantly enhancing its temporal forecasting capacity. Operationally, the Random Forest algorithm first identifies seismic events exceeding a magnitude of 5.0, after which the time-series Transformer predicts the exact occurrence window down to the minute. By combining these predictive outputs, the system dynamically generates visual warning alerts based on the forecasted severity and timeline. Empirical evaluations demonstrate strong predictive proficiency. The magnitude estimation model achieved a Mean Squared Error (MSE) of 0.0261, a Mean Absolute Error (MAE) of 0.0883, a Root Mean Squared Error (RMSE) of 0.1615, and a Coefficient of Determination (R^2) of 0.7737. For earthquake timing forecasting, the system yielded an MSE of 0.0007, an MAE of 0.0265, and an RMSE of 0.0264. Implementing this hybrid early warning system offers a vital tool for effective disaster risk management and mitigation strategies in seismically active regions.