Shagufta Henna
@atu.ie
Lecturer, Department of Computing
Atlantic Technological University, Letterkenny, Co. Donegal, Ireland
Shagufta Henna is Lecturer with the Department of Computing, Atlantic Technological University, Donegal, Ireland. She was a research fellow with the CONNECT, the Science Foundation Ireland Research Centre, 2019. She received her doctoral degree in Computer Science from the University of Leicester, UK, in 2013. She was an assistant professor from 2013 to 2018, Bahria University, Islamabad. She has published several scientific papers in leading journals/transactions and conferences. She has been involved in several EU and national research projects including H2020, Science Foundation of Ireland, and Enterprise of Ireland. She is a senior member of the IEEE, and is serving the editorial boards of IEEE Access, EURASIP Journal on Wireless Communications and Networking, IEEE Future Directions, and Springer Human-centric Computing and Information Sciences. She is currently supervising several PhD/Masters research students in wireless communication, deep learning, and big data analytics. She h
EDUCATION
PhD, Computer Science, University of Leicester, UK
RESEARCH INTERESTS
Deep Learning, Federated Learning, Explainable AI, Future Generation Networks, Edge Intelligence, Cybersecurity
Scopus Publications
Scopus Publications
Randika K. Makumbura, Lakindu Mampitiya, Namal Rathnayake, D.P.P. Meddage, Shagufta Henna, Tuan Linh Dang, Yukinobu Hoshino, and Upaka Rathnayake
Elsevier BV
Shagufta Henna, Stephen Azeez, Muhammad Bilal, and Aparna Reiji
AIP Publishing
Shagufta Henna, David Aboga, Muhammad Bilal, and Stephen Azeez
AIP Publishing
Mohamed Amjath and Shagufta Henna
IEEE
Preserving privacy is a critical challenge when applying federated learning (FL) to sensitive healthcare data in the Internet of Health Things (IoHT) applications. Despite employing a range of differential privacy (DP) analysis methods to assess privacy in FL, achieving an optimal balance between privacy and utility continues to be challenging. An incorrect balance in the privacy-utility tradeoff can significantly compromise patient privacy and affect the reputation of healthcare organizations. To address this issue, this research investigates a robust privacy-preserving framework utilising the Skellam mechanism and Renyi Differential Privacy (RDP) technique. By utilizing the closed-under summation property of the Skellam mechanism, we address the challenges in aggregating noisy updates in FL. Further, through a rigorous RDP analysis, we demonstrate that selecting an appropriate RDP order $a$ and privacy budget ∊ enables a balanced tradeoff within the FL setup. Finally, we validate our approach through the development of a novel algorithm, DP-DotGAT. The algorithm takes function call graphs (FCGs) obtained from healthcare applications as input. Additionally, it effectively protects against membership inference attacks while maintaining an accuracy range of 70% to 73%. Our results indicate that this approach greatly improves the privacy of health care data and can lead to wider adoption of secure and privacy-preserving machine learning models in healthcare sectors.
Asif Yar, Shagufta Henna, Marion McAfee, and Salem S. Gharbia
IEEE
Self-calibration poses one of the primary challenges in deploying wireless sensor networks (WSNs), particularly in uncontrolled environments. While existing deep learning methods have demonstrated their effectiveness in centralized scenarios, their efficiency diminishes when applied in distributed and uncontrolled environments. To tackle this issue, this work evaluates the application of extreme learning machines (ELM) and graph deep learning techniques for self-calibration. The experimental findings reveal that the graph learning approach outperforms state-of-the-art deep learning methods, including LSTM, MLP, and CNN, in terms of accuracy. Additionally, ELM showcases its effectiveness as a suitable machine learning approach under resource-constrained scenarios, with accelerated learning.
Mallikharjuna Rao Sakhamuri, Shagufta Henna, Leo Creedon, and Kevin Meehan
IEEE
An understanding of the human innate immune response is essential to accelerate the progress and clinical trials of drugs and antibiotics. This requires efficient modelling of the molecular network. A lack of diverse and vast molecular data has been a major hurdle in molecular modelling research. Existing synthetic graph models such as variational auto encoders, Restricted Boltzmann Machines and auto regressive models are computationally expensive and experience inefficiencies due to their sequential architectural design. To address these issues this research paper introduces MAGNet - a Molecular Adversarial Graph Network based graph generative AI model to generate synthetic graph structures. MAGNet model architecture consists of two neural networks, a generator and discriminator which are involved in a minmax game to improve each other's performance. The model is trained on the graphs generated from unlabelled SMILES strings. The MAGNet architecture enabled the model to converge in fewer epochs and the generator was able to produce an accurate synthetic graph which is indistinguishable by the discriminator. The experimental results establish that the Graph Adversarial Network model is able to converge the discriminator loss within the first 15 epochs of the training.
Stephen Azeez and Shagufta Henna
Springer Nature Singapore
Shagufta Henna, Abid Ali Minhas, Muhammad Saeed Khan, and Muhammad Shahid Iqbal
Elsevier BV
Mohamed Amjath and Shagufta Henna
IEEE
The rise of the Internet of Health Things (IoHTs) has resulted in a significant increase in collaborative initiatives among healthcare organizations employing federated learning (FL). Even though FL trains models locally to protect privacy, exchanging model parameters still creates privacy risks, especially when working with non-Euclidean data like graphs. To address this issue, differential privacy (DP) is widely used, however, choosing appropriate privacy parameters remains difficult. Therefore, this research employs Rényi Differential Privacy (RDP) analysis, which extends the capabilities of traditional DP by providing more flexibility in the selection of privacy parameters. To measure, this research first models the malware dataset as a function call graph (FCG). Subsequently, the DP-SGD-enabled DotGAT model is utilized to classify both malware and benign applications, ensuring the preservation of privacy while maintaining model utility. Finally, We empirically demonstrate that selecting Rényi divergence (a) values between 2 and 2.5 optimises the balance between privacy and utility in graph-based models within the FL setup, improving healthcare collaboration privacy.
Mallikharjuna Rao Sakhamuri, Shagufta Henna, Leo Creedon, and Kevin Meehan
IEEE
An understanding of the human innate immune re-sponse has the potential to accelerate the development and clinical trials of drugs and antibiotics. This includes an understanding of T-cell responses, peptides, and the intricate interactions with Human Leukocyte Antigens (HLA). Graph based models capture the structural aspects of these interactions efficiently and graph based neural networks are the best choice of tools to analyse these datasets. However, the polymorphic nature of peptides, coupled with various influencing factors, results in the representation of HLA-peptide interactions as large complex graphs. Traditional graph-based neural networks often face challenges in processing large graphs and suffer due to their high learning and training time with reduced generalization capabilities. To address these challenges, this paper proposes an influence maximization sub-graph sampling classification approach to retain the structural information of large complex molecular graphs. This is achieved using global-local influence maximization (GLIM) that combines Page Rank with Eigenvector centrality. This unique combination enables local substructure connectivity, considering edge weights, which are critical in molecular contexts. The graph neural network (GNN) classification model achieved 0.82% accuracy on the main graphs and 0.83% accuracy on the subgraphs. More notably, the model consumed approximately 81% less memory and was about 72% faster per epoch for the subgraphs compared to the main graphs. The experimental results demonstrate the effectiveness of the subgraphs as they minimize memory resource utilization and processing time, making them a practical and efficient choice for HLA-peptide immunogenetic behavior analysis.
Asif Yar, Shagufta Henna, Marion McAfee, and Salem S. Gharbia
IEEE
Calibrating wireless sensor network deployments, especially in uncontrolled environments, poses a significant challenge. Existing deep learning approaches for calibration perform well when resource requirements are not constrained However, conventional deep learning models are not well suited for resource constraint environments due to their computational complexity and requirement of resources. To tackle this issue, this paper introduces an extreme learning machine (ELM)-based calibration solution. ELM leverages a single-layer neural network with random weight initialization, enabling faster training and inference. Experimental results demonstrate that ELM results in accelerated learning while maintaining competitive accuracy compared to deep learning approaches like neural networks (NNs).
D.S. Wijenayake, Shagufta Henna, and William Farrelly
IEEE
When bidirectional, data-intensive, scientific applications move through multiple cloud environments, the posture requirements change during transit. Security is crucial in this scenario as the applications and as well as infrastructure providers demand it. Previous studies were based on sub-optimal heuristics, and machine learning with no consideration for the bidirectional nature of the data workflows. This limitation is addressed in this work by adopting the graph structure and predicting cloud providers based on the specifications. GraphSAGE performed well with an AUC of 0.94 while having a trade-off between performance and cost. The research outcomes highlight the applicability of graph-based solutions to security-aware cloud service provider selection for bidirectional multi-cloud data workflows.
Asif Yar, Shagufta Henna, Marion McAfee, and Salem S. Gharbia
IEEE
Air pollution, particularly high concentrations of Ozone (O3), poses a serious threat to human health and the environment. While deep learning algorithms have proven effective in air quality forecasting, current offline models struggle to capture the dynamic, time-evolving patterns generated by continuous air pollution monitoring data. Further, the time-consuming training process and computational demands hinder the practicality of these models. This paper presents a lightweight incremental learning model tailored for O3 forecasting. To evaluate its effectiveness, real data is employed and performance is evaluated using forecasting metrics and computational time. The results reveal that the incremental learning model surpasses the state-of-the-art model widely used in O3 and time series forecasting, demonstrating both superior accuracy and computational efficiency.
Kazeem Saheed and Shagufta Henna
IEEE
Recent cyberattacks have shifted their focus from causing financial loss or service disruption to covertly exfil-trating confidential data. Advanced Persistent Threats (APTs) pose a significant challenge due to their dynamic and sophis-ticated attack mechanisms. Unlike other cyberattacks, APTs are coordinated and targeted, executed by high-profile hackers who exploit identified vulnerabilities and deliver novel malware through phishing attacks to infiltrate networks. Traditional deep learning approaches for APT detection are static and lack adaptability, making them unsuitable for handling the dynamic and evolving attack scenarios commonly found in uncertain network traffic flows, such as multi-stage APT attacks. To address these challenges, this study proposes a Deep Reinforcement Learning approach for APT detection, referred to as APT-DRL. This approach dynamically learns from interactions with the environment, continuously adapting to emerging attack patterns. Performance evaluations demonstrate that APT-DRL effectively learns from dynamic network interactions, enabling it to formulate new policies for APT detection. Consequently, APT-DRL learns faster and achieves better accuracy compared to Feed Forward Neural Network (FNN) models, which lack the adaptability and learning capabilities of the proposed APT-DRL approach.
Shagufta Henna, Asif Yar, Kazeem Saheed, and Paulson Grigarichan
IEEE
Calibration in wireless sensor networks (WSNs) poses a significant challenge, particularly in uncontrolled environmental deployments for environmental monitoring, such as air pollution. Traditional calibration methods rely on centralized reference stations, which are costly to maintain, offer limited coverage, and calculate measurements as averages. However, with the rise of the Internet of Things (IoT), sensors present a cost-effective alternative for calibration compared to fixed reference stations. Nevertheless, in uncontrolled environments, sensors require self-recalibration to ensure accurate measurements for the reliable operation of WSNs without human intervention. Existing calibration approaches, such as LSTM, are computationally expensive, have higher memory requirements, and exhibit training instability, making them unsuitable for resource-constrained WSNs. This paper proposes a self-calibration approach for WSNs using the Gated Recurrent Unit (GRU) coupled with the attention mechanism (Attention-GRU). The Attention-GRU selectively focuses on relevant features while capturing long-term dependencies, akin to Recurrent Neural Networks (RNNs), thereby mitigating overfitting. Experimental results demonstrate that the Attention-GRU model outperforms other models with an R-squared value of 0.97 and accelerated learning. These accurate sensor recalibration predictions promote sustainability by supporting IoT-enabled air pollution monitoring efforts.
Kazeem Saheed and Shagufta Henna
IEEE
Advanced Persistent Threats (APTs) have significantly impacted organizations over an extended period with their coordinated and sophisticated cyberattacks. Unlike signature-based tools such as antivirus and firewalls that can detect and block other types of malware, APTs exploit zero-day vulnerabilities to generate new variants of undetectable malware. Additionally, APT adversaries engage in complex relationships and interactions within network entities, necessitating the learning of interactions in network traffic flows, such as hosts, users, or IP addresses, for effective detection. However, traditional deep neural networks often fail to capture the inherent graph structure and overlook crucial contextual information in network traffic flows. To address these issues, this research models APTs as heterogeneous graphs, capturing the diverse features and complex interactions in network flows. Consequently, a hetero-geneous graph transformer (HGT) model is used to accurately distinguish between benign and malicious network connections. Experiment results reveal that the HGT model achieves better performance, with 100 % accuracy and accelerated learning time, outperferming homogeneous graph neural network models.
Stephen Azeez and Shagufta Henna
Springer Nature Singapore
Mallikharjuna Rao Sakhamuri, Shagufta Henna, Leo Creedon, and Kevin Meehan
IEEE
The recent Covid-19 pandemic has attracted significant attention toward understanding the innate immune response of humans. To develop new drugs and antibiotics that activate T-cells to combat malicious pathogens, it is necessary to comprehend the immune system, including T-cell response, peptides, and Human Leukocyte Antigens (HLA) interactions. Traditional machine learning models, such as Convolutional Neural Networks (CNNs) and Feed Forward Networks (FNNs) are limited to feature extraction of peptides to predict immunogenicity values. CNNs and FNNs cannot capture the underlying structure and relationships between HLA and peptides, and therefore, do not assist with the immune response predictions. To address these issues, firstly this paper models the Immune Epitope dataset as a graph to capture dependencies and interactions among HLAs and peptides. Secondly, to assess the performance of the graph model, the results of the Graph Neural Network (GNN) are validated against the results of FNN. The results show that the GNN has better performance efficiency over conventional models in terms of accuracy and other performance metrics, thereby recommending graph-based deep learning as an efficient tool for drug discovery, diagnosis, and other immunology.
Shagufta Henna, Kevin Meehan, and Mallikharjuna Rao Sakhamuri
IEEE
Off-body communication in Cognitive Radio Wireless Body Area Networks (CRWBANs) must cope with different types of interference and dynamic traffic loads. Existing static channel assignment schemes cannot handle dynamic traffic loads with higher data rate requirements. One of the major problems experienced by CRWBANs is contention across wireless body area networks, thereby affecting the network throughput. Channel bonding has been successfully applied to different networks, such as Wireless Local Area Networks (WLANs), cognitive radio sensor networks, and wireless sensor networks. However, its use for the off-body communication in IEEE 802.15.6-based CRWBANs has not been investigated. This article proposes a multi-armed bandit-based bonded channel algorithm (MAB-BCA) with Upper Confidence Bound (UCB) for channel bonding to improve the off-body network capacity. The MAB-BCA based on UCB algorithm maximizes the capacity of off-body communication in CRWBANs. It demonstrates significant performance improvement over Static Channel Assignment (SCA) and Reinforcement Learning—Channel Assignment Algorithm (RL-CAA) in terms of average throughput and comparable bit error rate and dissatisfaction probability.
Mohamed Ibrahim, Shagufta Henna, and Gary Cullen
Springer Nature Switzerland
AbstractMammography is a popular diagnostic imaging procedure for detecting breast cancer at an early stage. Various deep-learning approaches to breast cancer detection incur high costs and are erroneous. Therefore, they are not reliable to be used by medical practitioners. Specifically, these approaches do not exploit complex texture patterns and interactions. These approaches warrant the need for labelled data to enable learning, limiting the scalability of these methods with insufficient labelled datasets. Further, these models lack generalisation capability to new-synthesised patterns/textures. To address these problems, in the first instance, we design a graph model to transform the mammogram images into a highly correlated multigraph that encodes rich structural relations and high-level texture features. Next, we integrate a pre-training self-supervised learning multigraph encoder (SSL-MG) to improve feature presentations, especially under limited labelled data constraints. Then, we design a semi-supervised mammogram multigraph convolution neural network downstream model (MMGCN) to perform multi-classifications of mammogram segments encoded in the multigraph nodes. Our proposed frameworks, SSL-MGCN and MMGCN, reduce the need for annotated data to 40% and 60%, respectively, in contrast to the conventional methods that require more than 80% of data to be labelled. Finally, we evaluate the classification performance of MMGCN independently and with integration with SSL-MG in a model called SSL-MMGCN over multi-training settings. Our evaluation results on DSSM, one of the recent public datasets, demonstrate the efficient learning performance of SSL-MNGCN and MMGCN with 0.97 and 0.98 AUC classification accuracy in contrast to the multitask deep graph (GCN) method Hao Du et al. (2021) with 0.81 AUC accuracy.
Zulfiqar Ali, Shagufta Henna, Adnan Akhunzada, Mohsin Raza, and Sung Won Kim
Institute of Electrical and Electronics Engineers (IEEE)
Shagufta Henna and Alan Davy
Institute of Electrical and Electronics Engineers (IEEE)
Ubiquitous computing has potentials to harness the flexibility of distributed computing systems including cloud, edge, and Internet of Things devices. Mobile edge computing (MEC) benefits time-critical applications by providing low latency connections. However, most of the resource-constrained edge devices are not computationally feasible to host deep learning (DL) solutions. Further, these edge devices if deployed under denser deployments result in topological dependencies which if not taken into consideration adversely affect the MEC performance. To bring more intelligence to the edge under topological dependencies, compared to optimization heuristics, this article proposes a novel collaborative distributed DL approach. The proposed approach exploits topological dependencies of the edge using a resource-optimized graph neural network (GNN) version with an accelerated inference. By exploiting edge collaborative learning using stochastic gradient (SGD), the proposed approach called CGNN-edge ensures fast convergence and high accuracy. Collaborative learning of the deployed CGNN-edge incurs extra communication overhead and latency. To cope, this article proposes compressed collaborative learning based on momentum correction called cCGNN-edge with better scalability while preserving accuracy. Performance evaluation under IEEE 802.11ax-high-density wireless local area networks deployment demonstrates that both the schemes outperform cloud-based GNN inference in response time, satisfaction of latency requirements, and communication overhead.
Abid Ali Minhas, Muhammad Saeed Khan, Shagufta Henna, and Muhammad Shahid Iqbal
SPIE-Intl Soc Optical Eng
Abstract. Due to high data rates, license-free spectrum, and immunity to electromagnetic interference, free-space optical (FSO) links are being considered as a potential candidate to meet the ever-increasing traffic demands of users. The FSO links remain less explored, as their performance depends on environmental conditions such as dust, fog, and clouds. Such conditions do not affect the radio frequency (RF) links in a similar way; however, RF resources are scarce. As an ultimate solution to this performance/scarcity dilemma, we propose a fuzzy logic-based hybrid architecture of FSO and RF links, which can be used to enhance reliability and resource efficiency. We present an intelligent soft switching mechanism between FSO and RF links using a fuzzy inference system to achieve the maximum link reliability and provide heterogeneous wireless services.
Shagufta Henna
Institute of Electrical and Electronics Engineers (IEEE)
Since the rapid outbreak of Covid-19, profound research interest has emerged to understand the innate immune response to viruses to enable appropriate vaccination. This understanding can help to inhibit virus replication, prolong adaptive immune response, accelerated virus clearance, and tissue recovery, a key milestone to combat coronaviruses (CoVs), e.g., Covid-19. An innate immune system triggers inflammatory responses against CoVs upon recognition of viruses. An appropriate defense against various coronavirus strains requires a deep understanding of the innate immune response system. Current deep learning approaches focus more on Covid-19 detection and pay no attention to understand the immune response once a virus invades. In this work, we propose a graph neural network-based (GNN) model that exploits the interactions between pattern recognition receptors (PRRs)to understand the human immune response system. PRRs are germline-encoded proteins that identify molecules related to pathogens and initiate a defense mechanism against the related pathogens, thereby aiding the innate immune response system. An understanding of PRR interactions can help to recognize pathogen-associated molecular patterns (PAMPs) to predict the activation requirements of each PRR. The immune response information of each PRR is derived from combining its historical PAMPs activation coupled with the modeled effect on the same from PRRs in its neighborhood. On one hand, this work can help to understand how long Covid-19 can confer immunity for a strong immune response. On the other hand, this GNN-based understanding can also abode well for appropriate vaccine development efforts against CoVs. Our proposal has been evaluated using CoVs immune response dataset, with results showing an average IFNs activation prediction accuracy of 90%, compared to 85% using feed-forward neural networks.
Azka Amin, Xi-Hua Liu, Muhammad Asim Saleem, Shagufta Henna, Taseer-ul Islam, Imran Khan, Peerapong Uthansakul, Muhammad Zeshan Qurashi, Seyed Sajad Mirjavadi, and Masoud Forsat
Hindawi Limited
Wireless power transfer techniques to transfer energy have been widely adopted by wireless rechargeable sensor networks (WRSNs). These techniques are aimed at increasing network lifetime by transferring power to end devices. Under these wireless techniques, the incurred charging latency to replenish the sensor nodes is considered as one of the major issues in wireless sensor networks (WSNs). Existing recharging schemes rely on rigid recharging schedules to recharge a WSN deployment using a single global charger. Although these schemes charge devices, they are not on-demand and incur higher charging latency affecting the lifetime of a WSN. This paper proposes a collaborative recharging technique to offload recharging workload to local chargers. Experiment results reveal that the proposed scheme maximizes average network lifetime and has better average charging throughput and charging latency compared to a global charger-based recharging.