Mohammed Elmusrati

The growing demand for intelligent and context-aware Internet of Things (IoT) systems has spurred the advancement of machine learning (ML) solutions for real-time localization. The deployment of ML models on distributed and resource-constrained edge devices offers substantial opportunities while posing unique challenges. This paper presents a structured and comparative review of strategies for adapting and designing ML models for deployment on embedded hardware with limited resources. We emphasize Tiny Machine Learning (TinyML), a paradigm that facilitates ultra-low-power ML execution on microcontrollers and sensors, and explore its synergy with Federated Learning (FL), which enables collaborative model training across devices without centralizing sensitive data. We examine training strategies, deployment workflows from cloud to edge, model optimization techniques, and key applications of TinyML and FL in localization. Benefits such as real-time processing, privacy preservation, and energy efficiency are analysed, along with challenges including computational constraints, interoperability, and security vulnerabilities. The article concludes by identifying research gaps and outlining future directions for integrating TinyML and FL in large-scale localization systems.

Modern cloud–edge computing environments host distributed applications whose resource consumption varies dynamically across multiple infrastructure layers. Traditional reactive monitoring and scaling mechanisms often fail to anticipate workload fluctuations, leading to performance degradation and inefficient resource allocation. Accurate prediction of application behavior is therefore essential for enabling proactive orchestration and intelligent resource management in cloud–edge systems.This paper proposes a hybrid machine learning framework for real-time prediction of application behavior across multiple performance metrics in the Cloud–Edge Computing Continuum (CECC). The framework integrates time-series feature engineering with ensemble tree-based learning models to forecast key resource utilization indicators, including CPU usage, memory consumption, disk throughput, and network traffic. By capturing temporal workload patterns and nonlinear dependencies among resource metrics, the proposed approach enables accurate short-term prediction suitable for operational environments.The proposed framework is evaluated using the large-scale GWA-T-13 Materna dataset, which contains production traces collected from more than 1,500 virtual machines running enterprise workloads. Experimental results demonstrate that the hybrid modeling approach consistently improves prediction accuracy compared with individual baseline models, including traditional statistical forecasting techniques and single-model machine learning approaches. The results show that combining ensemble learning methods with time-series feature extraction significantly enhances forecasting reliability for multi-metric application behavior.The findings highlight the potential of predictive analytics for enabling proactive orchestration strategies in cloud–edge platforms. By providing accurate forecasts of resource utilization, the proposed framework supports improved scheduling decisions, efficient resource allocation, and enhanced system reliability in distributed computing environments.

Extracting meaning from uncertain, noisy data is a fundamental problem across time series analysis, pattern recognition, and language modeling. This survey presents a unified mathematical framework that connects classical estimation theory, statistical inference, and modern machine learning, including deep learning and large language models. By analyzing how techniques such as maximum likelihood estimation, Bayesian inference, and attention mechanisms address uncertainty, the paper illustrates that many AI methods are rooted in shared probabilistic principles. Through illustrative scenarios including system identification, image classification, and language generation, we show how increasingly complex models build upon these foundations to tackle practical challenges like overfitting, data sparsity, and interpretability. In other words, the work demonstrates that maximum likelihood, MAP estimation, Bayesian classification, and deep learning all represent different facets of a shared goal: inferring hidden causes from noisy and/or biased observations. It serves as both a theoretical synthesis and a practical guide for students and researchers navigating the evolving landscape of machine learning.

The growth of satellite-based positioning methods has revolutionized global navigation by providing reliable geolocation capabilities. However, traditional Global Navigation Satellite Systems (GNSS) are increasingly vulnerable to threats like jamming, spoofing, and interception, undermining their reliability in critical applications such as in-flight navigation and emergency services. To address these challenges, Low Earth Orbit (LEO) satellite constellations have emerged as a promising complement to GNSS infrastructure. LEO satellites, orbiting at lower altitudes with higher density, offer improved signal availability, reduced degradation, and better reception on Earth. This paper presents a LEO satellite-based positioning method via massive multiple-input multiple-output (mMIMO) beamforming antennas. The proposed technique not only mitigates GNSS vulnerabilities but also introduces a passive sensing mechanism that facilitates positioning without complex timing synchronization, improving resilience in jamming-prone environments. By utilizing LEO satellite beam identifiers as geographic pointers, our method enables precise positioning through LEO satellite ephemeris and beam pattern data. We validate this beam-based method through simulations, LEO constellation data, vehicular drive-test datasets, and probabilistic positioning models. Positioning results from the first dataset show a mean absolute error (MAE) of 9.15 meters and a 95th percentile error (p95%) of 19.07 meters when combining LEO satellite data with inertial motion data from a moving vehicle. Meanwhile, GNSS accuracy was MAE = 26.6 meters and p95% = 56.6 meters. The second dataset showed consistent results with accuracy improvements in MAE from 18.55 to 9.42 meters, RMSE from 22.24 to 12.05 meters, and p95% from 36.38 to 21.18 meters, compared to GNSS. These findings highlight the potential of LEO satellite positioning to improve accuracy and reliability in challenging environments, with implications for critical applications such as remote sensing, emergency response, search and rescue, and situational awareness.

There is a lack of prognosticators of overall survival (OS) for Oral Squamous Cell Carcinoma (OSCC). We examined collaborative machine learning (cML) in estimating the OS of OSCC patients. The prognostic significance of the clinicopathological parameters was examined. Altogether, 9439 OSCC patients were extracted from the Surveillance, Epidemiology, and End Results database (US). Five ML models - voting ensemble, stacked ensemble, extreme gradient boosting, light boosting, and logistic regression were used to predict OS. Three of these ML algorithms were combined to form a cluster of cML models. The performance of the cML was compared with the best performing individual ML algorithm following model training. The performance accuracy of the voting ensemble, stacked ensemble, extreme gradient boosting, light boosting, and logistic regression models was 70.2%, 69.9%, 69.1%, 69.4%, and 69.5% respectively, following model training. When the voting ensemble model was compared with cML using temporal validation, the cML showed a comparable performance accuracy. The most significant prognostic factors were age of the patient at diagnosis, T stage, tumor grade, marital status, gender, primary site, surgery, N stage, radiotherapy, ethnicity, chemotherapy, and M stage. cML appears to give reliability to the final prediction and thereby may mark a paradigm shift from model individualism to a more cooperative paradigm. This approach may aid in determining an enhanced individualized treatment for OSCC patients.

Designing a globally ubiquitous positioning system that works seamlessly in all environments remains a critical area of ongoing research. While Global Navigation Satellite Systems (GNSS), such as GPS, are the predominant technology for global outdoor positioning, they fail in areas with physical obstructions (e.g., dense urban regions and indoors) and are prone to jamming. These limitations significantly restrict their global accessibility and dependability, highlighting the need for supplementary positioning technologies. Recently, private companies like SpaceX have started deploying Low Earth Orbit (LEO) satellites for various applications, primarily communication. LEO satellites operate at much lower altitudes than GNSS satellites, offering higher signal penetration capabilities and reduced susceptibility to jamming, making them a promising complementary positioning technology. In this paper, we present our vision for opportunistically leveraging LEO satellite signals to achieve ubiquitous and reliable global positioning. We explore diverse opportunities enabled by our vision and discuss the multi-disciplinary challenges that must be addressed for its realization.