Mohamed El Ghmary
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Mohamed El Ghmary has a total of 45 co-authors. Here are the top co-authors by number of shared publications:
- Mohamed Amnai
Ibn Tofail University
9 shared publications - Said Ziani
Mohammed V University
7 shared publications - Sara Maftah
Ibn Tofail University
7 shared publications - Abderrahim Ouza
Ibn Tofail University
6 shared publications - Ali Choukri
Ibn Tofail University
6 shared publications - Ali Ouacha
Mohammed V University in Rabat
6 shared publications
Latest Publications
- 23 September 2026
- Published by Springer Nature in Multimedia Tools and Applications
- Vol. 85 (10) , 775
Hybrid convolutional neural network with attention mechanism to provide robust face expression recognition
- 23 September 2026
- Published by Springer Nature in Multimedia Tools and Applications
- Vol. 85 (10) , 775
Hybrid convolutional neural network with attention mechanism to provide robust face expression recognition
- Open Access
- 21 September 2026
- Published by Elsevier in SSRN Electronic Journal
A hybrid machine learning framework for energy-aware virtual machinescheduling in cloud computing
Energy efficiency in cloud data centres is a critical operational and environmental challenge, yet widely deployed virtual machine (VM) schedulers such as Round Robin, First-Fit Decreasing, and Least-CPU lack explicit energy awareness. This paper introduces Machine Learning Energy-Aware (ML-EA), a hybrid VM scheduler that integrates empirical server power models with supervised regression learning in a closed-loop assignment mechanism. Three physics-based power models (M1–M3) generate regression labels by computing the incremental energy, in watt-seconds (W·s), of each task–VM pairing; these physics-derived labels serve as the ground truth for supervised learning. Among the three formulations—linear CPU (M1), multi-component additive (M2), and polynomial nonlinear (M3)—M3 is paired with XGBoost as the primary scheduler, as it provides the sharpest energy discrimination between VM placements. The pipeline is trained on the Google Cluster Trace v3 (GCT-v3) and validated through 5-fold cross-validation, achieving R2 = 0.999 ± 0.001 andMAE = 4.38 ± 0.15 W·s. Evaluated on the out-of-distribution Alibaba Cluster Trace v2 (ACT-v2) without retraining, the same pipeline reaches R2 = 0.9981 ± 0.0012 and MAE = 5.14 ± 0.18 W·s. Integrated into an online scheduling loop, ML-EA reduces cumulative energy consumption by up to 72.2 % versus Round Robin on GCT-v3 and 68.4 % on ACT-v2, across workload sizes of 300–1500 tasks, with scheduling overhead below 25 ms per task.
- 21 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA hybrid machine learning framework for energy-aware virtual machinescheduling in cloud computing
Energy efficiency in cloud data centres is a critical operational and environmental challenge, yet widely deployed virtual machine (VM) schedulers such as Round Robin, First-Fit Decreasing, and Least-CPU lack explicit energy awareness. This paper introduces Machine Learning Energy-Aware (ML-EA), a hybrid VM scheduler that integrates empirical server power models with supervised regression learning in a closed-loop assignment mechanism. Three physics-based power models (M1–M3) generate regression labels by computing the incremental energy, in watt-seconds (W·s), of each task–VM pairing; these physics-derived labels serve as the ground truth for supervised learning. Among the three formulations—linear CPU (M1), multi-component additive (M2), and polynomial nonlinear (M3)—M3 is paired with XGBoost as the primary scheduler, as it provides the sharpest energy discrimination between VM placements. The pipeline is trained on the Google Cluster Trace v3 (GCT-v3) and validated through 5-fold cross-validation, achieving R2 = 0.999 ± 0.001 andMAE = 4.38 ± 0.15 W·s. Evaluated on the out-of-distribution Alibaba Cluster Trace v2 (ACT-v2) without retraining, the same pipeline reaches R2 = 0.9981 ± 0.0012 and MAE = 5.14 ± 0.18 W·s. Integrated into an online scheduling loop, ML-EA reduces cumulative energy consumption by up to 72.2 % versus Round Robin on GCT-v3 and 68.4 % on ACT-v2, across workload sizes of 300–1500 tasks, with scheduling overhead below 25 ms per task.
- Open Access
- 6 August 2026
- Published by International Association of Online Engineering (IAOE) in International Journal of Interactive Mobile Technologies (iJIM)
- Vol. 20 (15) , 4–21
Benchmarking the Pareto Frontier of VM Placement: A Multi-Objective Evaluation of Heuristics and Metaheuristics Algorithms
Virtual machine (VM) placement in recent cloud systems remains a challenging and new task due to the challenge of finding the best way to use resources efficiently while also ensuring service reliability. The current study offers an in-depth comparative evaluation of six VM placement algorithms—three established heuristics (Best Fit Decreasing (BFD), First Fit Decreasing (FFD), and Worst Fit (WF)), a baseline RANDOM method, and two metaheuristics (Ant Colony Optimization (ACO) and NSGA-III)—conducted through 2,400 simulation trials. We meticulously evaluate each algorithm across four critical performance metrics: placement success rate, energy consumption, execution time, and service reliability. The statistical analysis confirmed significant differences in performance across all methods using the Kruskal-Wallis and Mann-Whitney tests. Classical heuristics achieve a placement success rate (PSR) of 87.38% with BFD and FFD operating without any service violations. These methods use more energy than other approaches, which is a drawback of their good performance. NSGA-III reduces energy usage by 11% while maintaining acceptable placement performance (71.62%). The execution time (ET) can really vary, from super-fast heuristics to more time-consuming optimization methods. The results indicate that classical heuristics provide the highest reliability. No SLA violations (SLAV) were observed in the scenarios evaluated. Classical heuristics, in particular BFD and FFD, are therefore the most reliable. The study’s findings also show that cloud providers can choose the algorithms that best match their operational priorities. These priorities may include maximizing performance, minimizing energy consumption, or ensuring service stability.
- 6 August 2026
- Published by International Association of Online Engineering (IAOE) in International Journal of Interactive Mobile Technologies (iJIM)
- Vol. 20 (15) , 4–21
Open AccessBenchmarking the Pareto Frontier of VM Placement: A Multi-Objective Evaluation of Heuristics and Metaheuristics Algorithms
Virtual machine (VM) placement in recent cloud systems remains a challenging and new task due to the challenge of finding the best way to use resources efficiently while also ensuring service reliability. The current study offers an in-depth comparative evaluation of six VM placement algorithms—three established heuristics (Best Fit Decreasing (BFD), First Fit Decreasing (FFD), and Worst Fit (WF)), a baseline RANDOM method, and two metaheuristics (Ant Colony Optimization (ACO) and NSGA-III)—conducted through 2,400 simulation trials. We meticulously evaluate each algorithm across four critical performance metrics: placement success rate, energy consumption, execution time, and service reliability. The statistical analysis confirmed significant differences in performance across all methods using the Kruskal-Wallis and Mann-Whitney tests. Classical heuristics achieve a placement success rate (PSR) of 87.38% with BFD and FFD operating without any service violations. These methods use more energy than other approaches, which is a drawback of their good performance. NSGA-III reduces energy usage by 11% while maintaining acceptable placement performance (71.62%). The execution time (ET) can really vary, from super-fast heuristics to more time-consuming optimization methods. The results indicate that classical heuristics provide the highest reliability. No SLA violations (SLAV) were observed in the scenarios evaluated. Classical heuristics, in particular BFD and FFD, are therefore the most reliable. The study’s findings also show that cloud providers can choose the algorithms that best match their operational priorities. These priorities may include maximizing performance, minimizing energy consumption, or ensuring service stability.
- Open Access
- 6 August 2026
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 6 (1)
Resource-aware hybrid homomorphic encryption scheduling for privacy-preserving edge intelligence
- 6 August 2026
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 6 (1)
Open AccessResource-aware hybrid homomorphic encryption scheduling for privacy-preserving edge intelligence
- 2 August 2026
- Published by Springer Nature
- p. 75-82
Intelligent Continuous Deep Learning System for Stereo Camera Self-calibration
- 2 August 2026
- Published by Springer Nature
- p. 75-82
Intelligent Continuous Deep Learning System for Stereo Camera Self-calibration
- 2 August 2026
- Published by Springer Nature
- p. 607-613
A Comparative Study of License Plate Localization Using Artificial Intelligence Techniques
- 2 August 2026
- Published by Springer Nature
- p. 607-613
A Comparative Study of License Plate Localization Using Artificial Intelligence Techniques
- 2 August 2026
- Published by Springer Nature
- p. 172-178
Enhancing Non-invasive Fetal Electrocardiogram Extraction Through Subspace Projection and Neural Demixing
- 2 August 2026
- Published by Springer Nature
- p. 172-178
Enhancing Non-invasive Fetal Electrocardiogram Extraction Through Subspace Projection and Neural Demixing
- 2 August 2026
- Published by Springer Nature
- p. 463-469
Predicting Social Crises and Territorial Instability Using Machine Learning: A Framework for Proactive Governance
- 2 August 2026
- Published by Springer Nature
- p. 463-469
Predicting Social Crises and Territorial Instability Using Machine Learning: A Framework for Proactive Governance