Ghada Alsuhli
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Ghada Alsuhli has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
- Hani Saleh
Khalifa University
6 shared publications - Thanos Stouraitis
Department of Computer and Information Engineering, System on Chip Center, Khalifa University, Abu Dhabi, United Arab Emirates
5 shared publications - Baker Mohammad
Khalifa University, Abu Dhabi, United Arab Emirates
3 shared publications - Mahmoud Al-Qutayri
Department of Computer and Information Engineering, System on Chip Center, Khalifa University, Abu Dhabi, United Arab Emirates
3 shared publications Mahmoud Al-Qutayri
2 shared publicationsThanos Stouraitis
2 shared publications
Latest Publications
- Open Access
- 22 May 2026
- Published by Elsevier in Microprocessors and Microsystems
- Vol. 122
Design and implementation of a fixed-point FFT/IFFT processor for NTRU lattices
- 22 May 2026
- Published by Elsevier in Microprocessors and Microsystems
- Vol. 122
Open AccessDesign and implementation of a fixed-point FFT/IFFT processor for NTRU lattices
- Open Access
- 30 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (9) , 18839-18856
Casting Ventricular Arrhythmia Detection as Anomaly Detection via One-Class Meta-Learning
Ventricular arrhythmia (VA) detection is a critical yet challenging task in cardiac healthcare due to the rarity of abnormal episodes and the high interpatient variability in cardiac signals. These challenges are further exacerbated in implantable cardioverter-defibrillators (ICDs), which operate under stringent memory and computational constraints. In this article, we analyze inter and intrapatient variability using dimensionality reduction and divergence metrics, and leverage these observations to formulate VA detection as a deployment-aligned one-class meta-learning problem. Accordingly, we adopt a one-class model-agnostic meta-learning (OC-MAML) with a clinically grounded task design that reflects real-world deployment conditions. Specifically, patient-disjoint support and query sets are used to simulate realistic distribution shifts and interpatient variability. By training primarily on normal intracardiac electrogram (IEGM) segments, the OC-MAML-based framework learns a task-agnostic initialization that rapidly adapts to new patients using only a few normal samples, thereby substantially reducing dependence on labeled arrhythmic data. Compared to conventional vanilla model-agnostic meta-learning (MAML), our proposed OC-MAML-based framework achieves a relative improvement of +14.1% in sensitivity, +2.5% in balanced accuracy, and +4.1% in $F_{1}$ -score, while reducing adaptation time by $5\times $ and maintaining comparable memory efficiency. These results underscore the framework’s potential for scalable, nearly label-free deployment in edge-based cardiac monitoring systems. The code for the proposed framework is publicly available at https://github.com/jaradat/VAD-OC-MAML
- 30 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (9) , 18839-18856
Open AccessCasting Ventricular Arrhythmia Detection as Anomaly Detection via One-Class Meta-Learning
Ventricular arrhythmia (VA) detection is a critical yet challenging task in cardiac healthcare due to the rarity of abnormal episodes and the high interpatient variability in cardiac signals. These challenges are further exacerbated in implantable cardioverter-defibrillators (ICDs), which operate under stringent memory and computational constraints. In this article, we analyze inter and intrapatient variability using dimensionality reduction and divergence metrics, and leverage these observations to formulate VA detection as a deployment-aligned one-class meta-learning problem. Accordingly, we adopt a one-class model-agnostic meta-learning (OC-MAML) with a clinically grounded task design that reflects real-world deployment conditions. Specifically, patient-disjoint support and query sets are used to simulate realistic distribution shifts and interpatient variability. By training primarily on normal intracardiac electrogram (IEGM) segments, the OC-MAML-based framework learns a task-agnostic initialization that rapidly adapts to new patients using only a few normal samples, thereby substantially reducing dependence on labeled arrhythmic data. Compared to conventional vanilla model-agnostic meta-learning (MAML), our proposed OC-MAML-based framework achieves a relative improvement of +14.1% in sensitivity, +2.5% in balanced accuracy, and +4.1% in $F_{1}$ -score, while reducing adaptation time by $5\times $ and maintaining comparable memory efficiency. These results underscore the framework’s potential for scalable, nearly label-free deployment in edge-based cardiac monitoring systems. The code for the proposed framework is publicly available at https://github.com/jaradat/VAD-OC-MAML
- Open Access
- 18 August 2025
- Published by Elsevier in Journal of Information Security and Applications
- Vol. 93
Efficient NTT/INTT processor for FALCON post-quantum cryptography
- 18 August 2025
- Published by Elsevier in Journal of Information Security and Applications
- Vol. 93
Open AccessEfficient NTT/INTT processor for FALCON post-quantum cryptography
- Open Access
- 26 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Proceedings of the IEEE
- Vol. 113 (2) , 172-207
A Survey and Comparative Analysis of Number Systems for Deep Neural Networks
Deep neural networks (DNNs) are indispensable in various artificial intelligence (AI) applications. However, their inherent complexity presents significant challenges, particularly when deploying them on resource-constrained devices. To overcome these hurdles, academia and industry are actively seeking ways to accelerate and optimize DNN implementations. A significant area of research revolves around discovering more effective methods to represent the enormous data volumes processed by DNNs. Traditional number systems (NSs) have proven nonoptimal for this task, prompting extensive exploration into alternative and bespoke systems for DNNs. This survey aims to comprehensively discuss various NSs utilized to efficiently represent DNN data. These systems are categorized mainly based on their impact on DNN performance and hardware implementation. This survey offers an overview of these categorized NSs and delves into different subsystems within each, outlining their effect on DNN performance and hardware design. Furthermore, these systems are compared quantitatively and qualitatively concerning their expected quantization error, memory utilization, and computational requirements. This survey also emphasizes the challenges linked with each system and the diverse proposed solutions to address them. Insights into the utilization of these NSs for sophisticated DNNs are also presented in this survey. Readers will acquire a deeper understanding of the importance of efficient NSs for DNNs, explore commonly used systems, comprehend the tradeoffs between these systems, delve into design considerations influencing their impact on DNN performance, and discover recent trends and potential research avenues in this field.
- 26 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Proceedings of the IEEE
- Vol. 113 (2) , 172-207
Open AccessA Survey and Comparative Analysis of Number Systems for Deep Neural Networks
Deep neural networks (DNNs) are indispensable in various artificial intelligence (AI) applications. However, their inherent complexity presents significant challenges, particularly when deploying them on resource-constrained devices. To overcome these hurdles, academia and industry are actively seeking ways to accelerate and optimize DNN implementations. A significant area of research revolves around discovering more effective methods to represent the enormous data volumes processed by DNNs. Traditional number systems (NSs) have proven nonoptimal for this task, prompting extensive exploration into alternative and bespoke systems for DNNs. This survey aims to comprehensively discuss various NSs utilized to efficiently represent DNN data. These systems are categorized mainly based on their impact on DNN performance and hardware implementation. This survey offers an overview of these categorized NSs and delves into different subsystems within each, outlining their effect on DNN performance and hardware design. Furthermore, these systems are compared quantitatively and qualitatively concerning their expected quantization error, memory utilization, and computational requirements. This survey also emphasizes the challenges linked with each system and the diverse proposed solutions to address them. Insights into the utilization of these NSs for sophisticated DNNs are also presented in this survey. Readers will acquire a deeper understanding of the importance of efficient NSs for DNNs, explore commonly used systems, comprehend the tradeoffs between these systems, delve into design considerations influencing their impact on DNN performance, and discover recent trends and potential research avenues in this field.
- 23 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 6 (12) , 3273-3286
Efficient Transformer Inference Through Hybrid Dynamic Pruning
In the world of deep learning, transformer models have become very significant, leading to improvements in many areas from understanding language to recognizing images, covering a wide range of applications. Despite their success, the deployment of these models in real-time applications, particularly on edge devices, poses significant challenges due to their computational intensity and memory demands. To overcome these challenges, we introduce a novel Hybrid Dynamic Pruning (HDP) technique, an efficient algorithm-architecture co-design approach that accelerates transformers using head sparsity, block sparsity and approximation to reduce computations in attention and reduce memory access. With the observation of the huge redundancy in attention scores and attention heads, we propose a novel integer-based block pruning to prune unimportant blocks in the attention matrix at run time. We also propose integerbased head pruning to detect and prune unimportant heads at an early stage at run time. Also, we propose an approximation method that reduces attention computations. To efficiently support these methods with lower latency, we propose the HDP Accelerator (HDPA) as a co-processor architecture, synthesized in two configurations—HDPA-edge and HDPA-server—to meet the needs of mobile and server platforms. Extensive experiments with different transformer models and benchmarks demonstrate that HDPA-server achieves 481× and 381× speedup in attention layer computation over Intel i7-1185G7 CPU and NVIDIA T4 GPU, respectively. Compared to other state-of-the-art accelerators, HDPA achieves 1.26× to 2.08× higher throughput, 1.3× to 18× greater MAC efficiency, and 1.1× to 5.1× improved energy efficiency, when normalized to the same computational load.
- 23 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 6 (12) , 3273-3286
Efficient Transformer Inference Through Hybrid Dynamic Pruning
In the world of deep learning, transformer models have become very significant, leading to improvements in many areas from understanding language to recognizing images, covering a wide range of applications. Despite their success, the deployment of these models in real-time applications, particularly on edge devices, poses significant challenges due to their computational intensity and memory demands. To overcome these challenges, we introduce a novel Hybrid Dynamic Pruning (HDP) technique, an efficient algorithm-architecture co-design approach that accelerates transformers using head sparsity, block sparsity and approximation to reduce computations in attention and reduce memory access. With the observation of the huge redundancy in attention scores and attention heads, we propose a novel integer-based block pruning to prune unimportant blocks in the attention matrix at run time. We also propose integerbased head pruning to detect and prune unimportant heads at an early stage at run time. Also, we propose an approximation method that reduces attention computations. To efficiently support these methods with lower latency, we propose the HDP Accelerator (HDPA) as a co-processor architecture, synthesized in two configurations—HDPA-edge and HDPA-server—to meet the needs of mobile and server platforms. Extensive experiments with different transformer models and benchmarks demonstrate that HDPA-server achieves 481× and 381× speedup in attention layer computation over Intel i7-1185G7 CPU and NVIDIA T4 GPU, respectively. Compared to other state-of-the-art accelerators, HDPA achieves 1.26× to 2.08× higher throughput, 1.3× to 18× greater MAC efficiency, and 1.1× to 5.1× improved energy efficiency, when normalized to the same computational load.
- 7 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computing
- Vol. 13 (2) , 423-437
Area and Power Efficient FFT/IFFT Processor for FALCON Post-Quantum Cryptography
Quantum computing is an emerging technology on the verge of reshaping industries, while simultaneously challenging existing cryptographic algorithms. FALCON, a recent standard quantum-resistant digital signature, presents a challenging hardware implementation due to its extensive non-integer polynomial operations, necessitating FFT over the ring Q[x]/(x n+1). This paper introduces an ultra-low-power and compact processor tailored for FFT/IFFT operations over the ring for efficient FALCON implementation. The proposed processor incorporates various optimization techniques, including twiddle factor compression and conflict-free scheduling. In an ASIC implementation using a 22 nm GF process, the proposed processor demonstrates an area occupancy of 0.15 mm2 and a power consumption of 12.6 mW/28.1 mW at an operating frequency of 167 MHz/500 MHz for the non-pipelined/pipelined version of the processor. Since a hardware implementation of FFT/IFFT over the ring is currently non-existent, the execution time achieved by this processor is compared to the reference software implementation of FFT/IFFT of FALCON on a Raspberry Pi 4 with Cortex-A72, where the proposed pipelined processor achieves a speedup up to 3.8×. Furthermore, in comparison to dedicated state-of-the-art hardware accelerators for classic FFT, the pipelined architecture occupies 42% less area and consumes 64% less power, on average. The quantified speedup in the context of FALCON suggests that the proposed hardware design offers a promising solution for the efficient implementation of FALCON.
- 7 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computing
- Vol. 13 (2) , 423-437
Area and Power Efficient FFT/IFFT Processor for FALCON Post-Quantum Cryptography
Quantum computing is an emerging technology on the verge of reshaping industries, while simultaneously challenging existing cryptographic algorithms. FALCON, a recent standard quantum-resistant digital signature, presents a challenging hardware implementation due to its extensive non-integer polynomial operations, necessitating FFT over the ring Q[x]/(x n+1). This paper introduces an ultra-low-power and compact processor tailored for FFT/IFFT operations over the ring for efficient FALCON implementation. The proposed processor incorporates various optimization techniques, including twiddle factor compression and conflict-free scheduling. In an ASIC implementation using a 22 nm GF process, the proposed processor demonstrates an area occupancy of 0.15 mm2 and a power consumption of 12.6 mW/28.1 mW at an operating frequency of 167 MHz/500 MHz for the non-pipelined/pipelined version of the processor. Since a hardware implementation of FFT/IFFT over the ring is currently non-existent, the execution time achieved by this processor is compared to the reference software implementation of FFT/IFFT of FALCON on a Raspberry Pi 4 with Cortex-A72, where the proposed pipelined processor achieves a speedup up to 3.8×. Furthermore, in comparison to dedicated state-of-the-art hardware accelerators for classic FFT, the pipelined architecture occupies 42% less area and consumes 64% less power, on average. The quantified speedup in the context of FALCON suggests that the proposed hardware design offers a promising solution for the efficient implementation of FALCON.
- Open Access
- 1 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Efficient Twiddle Factor Generation for Post Quantum Cryptography FALCON-based Number Theoretic Transform
Area and power-efficient hardware implementations are crucial for the widespread adoption of post-quantum cryptography (PQC) algorithms like FALCON. One of the main operations in FALCON is the Number Theoretic Transform (NTT), which needs to be performed with many prime numbers. Having one set of twiddle factors (TFs) for each prime makes storing all of these TFs impractical. In this paper, we propose an architecture for generating TFs on the fly for FALCON-oriented NTT, designed for area and power efficiency. Our approach dynamically generates TFs during NTT computations, significantly reducing on-chip memory requirements. The ASIC implementation results demonstrate significant improvements, with the proposed design reducing on-chip memory requirements by 99%, occupying 95% less area, and consuming 87.4% less power compared to the traditional ROM-based implementation. Furthermore, our design achieved a much higher maximum clock frequency, indicating superior performance in accessing twiddle factors. These findings highlight the potential of our proposed architecture for efficient hardware implementations of FALCON-based cryptographic systems.
- 1 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessEfficient Twiddle Factor Generation for Post Quantum Cryptography FALCON-based Number Theoretic Transform
Area and power-efficient hardware implementations are crucial for the widespread adoption of post-quantum cryptography (PQC) algorithms like FALCON. One of the main operations in FALCON is the Number Theoretic Transform (NTT), which needs to be performed with many prime numbers. Having one set of twiddle factors (TFs) for each prime makes storing all of these TFs impractical. In this paper, we propose an architecture for generating TFs on the fly for FALCON-oriented NTT, designed for area and power efficiency. Our approach dynamically generates TFs during NTT computations, significantly reducing on-chip memory requirements. The ASIC implementation results demonstrate significant improvements, with the proposed design reducing on-chip memory requirements by 99%, occupying 95% less area, and consuming 87.4% less power compared to the traditional ROM-based implementation. Furthermore, our design achieved a much higher maximum clock frequency, indicating superior performance in accessing twiddle factors. These findings highlight the potential of our proposed architecture for efficient hardware implementations of FALCON-based cryptographic systems.
- 9 January 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Deep Reinforcement Learning-based CIO and Energy Control for LTE Mobility Load Balancing
cellular networks' congestion has been one of the most common problems in cellular networks due to the huge increase in network load resulted from enhancing communication quality as well as increasing the number of users. Since mobile users are not uniformly distributed in the network, the need for load balancing as a cellular networks' self-optimization technique has increased recently. Then, the congestion problem can be handled by evenly distributing the network load among the network resources. Lots of research has been dedicated to developing load balancing models for cellular networks. Most of these models rely on adjusting the Cell Individual Offset (CIO) parameters which are designed for self-optimization techniques in cellular networks. In this paper, a new deep reinforcement learning-based load balancing approach is proposed as a solution for the LTE Downlink congestion problem. This approach does not rely only on adapting the CIO parameters, but it rather has two degrees of control; the first one is adjusting the CIO parameters, and the second is adjusting the eNodeBs' transmission power. The proposed model uses Double Deep Q-Network (DDQN) to learn how to adjust these parameters so that a better load distribution in the overall network is achieved. Simulation results prove the effectiveness of the proposed approach by improving the network overall throughput by up to 21.4% and 6.5% compared to the base-line scheme and the scheme that only adapts CIOs, respectively.
- 9 January 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Deep Reinforcement Learning-based CIO and Energy Control for LTE Mobility Load Balancing
cellular networks' congestion has been one of the most common problems in cellular networks due to the huge increase in network load resulted from enhancing communication quality as well as increasing the number of users. Since mobile users are not uniformly distributed in the network, the need for load balancing as a cellular networks' self-optimization technique has increased recently. Then, the congestion problem can be handled by evenly distributing the network load among the network resources. Lots of research has been dedicated to developing load balancing models for cellular networks. Most of these models rely on adjusting the Cell Individual Offset (CIO) parameters which are designed for self-optimization techniques in cellular networks. In this paper, a new deep reinforcement learning-based load balancing approach is proposed as a solution for the LTE Downlink congestion problem. This approach does not rely only on adapting the CIO parameters, but it rather has two degrees of control; the first one is adjusting the CIO parameters, and the second is adjusting the eNodeBs' transmission power. The proposed model uses Double Deep Q-Network (DDQN) to learn how to adjust these parameters so that a better load distribution in the overall network is achieved. Simulation results prove the effectiveness of the proposed approach by improving the network overall throughput by up to 21.4% and 6.5% compared to the base-line scheme and the scheme that only adapts CIOs, respectively.
- Open Access
- 1 October 2020
- Published by Institution of Engineering and Technology (IET) in IET Intelligent Transport Systems
- Vol. 14 (10) , 1190-1199
Bio‐inspired metaheuristic framework for clustering optimisation in VANETs
Vehicular ad‐hoc network (VANET) is a key enabling technology of intelligent transportation systems. VANETs are characterised by the rapidly changing topology and the unbounded network size. These characteristics present a range of challenges to different VANET applications such as routing and security. Clustering has strongly presented itself as an efficient solution to such challenges. In this study, the authors formulate the clustering algorithm as a many‐objective optimisation problem. Then, they propose a unified framework to optimise the configuration parameters arbitrary clustering algorithms. Three many‐objective metaheuristic optimisation techniques, ESPEA, MOEA/DD and NSGA‐III, are compared in context of this framework, and various commonly used quality indicators are utilised to identify the metaheuristic with the best quality of solutions. The proposed framework is then used to optimise a recent clustering algorithm. Using the optimal configuration resulting from the proposed framework significantly improves the performance of the clustering algorithm under‐test compared to the non‐optimised algorithm as well as other clustering approaches. This is demonstrated by the simulation results which showed up to 182% improvement in the cluster head lifetime and a reduction of 36% in the clustering packets overhead in the highway environment.
- 1 October 2020
- Published by Institution of Engineering and Technology (IET) in IET Intelligent Transport Systems
- Vol. 14 (10) , 1190-1199
Open AccessBio‐inspired metaheuristic framework for clustering optimisation in VANETs
Vehicular ad‐hoc network (VANET) is a key enabling technology of intelligent transportation systems. VANETs are characterised by the rapidly changing topology and the unbounded network size. These characteristics present a range of challenges to different VANET applications such as routing and security. Clustering has strongly presented itself as an efficient solution to such challenges. In this study, the authors formulate the clustering algorithm as a many‐objective optimisation problem. Then, they propose a unified framework to optimise the configuration parameters arbitrary clustering algorithms. Three many‐objective metaheuristic optimisation techniques, ESPEA, MOEA/DD and NSGA‐III, are compared in context of this framework, and various commonly used quality indicators are utilised to identify the metaheuristic with the best quality of solutions. The proposed framework is then used to optimise a recent clustering algorithm. Using the optimal configuration resulting from the proposed framework significantly improves the performance of the clustering algorithm under‐test compared to the non‐optimised algorithm as well as other clustering approaches. This is demonstrated by the simulation results which showed up to 182% improvement in the cluster head lifetime and a reduction of 36% in the clustering packets overhead in the highway environment.