Mohammed Kaosar

Topic modeling is an unsupervised technique in natural language processing (NLP) used to identify hidden topic structures within large text datasets. Among traditional approaches to topic modeling, latent Dirichlet allocation, BERTopic, and Top2Vec, are widely adopted to uncover hidden topics in text data. However, these methods often struggle with poor performance in scenarios involving limited data availability or high-dimensional textual features. In this research, we propose QTopic, a novel hybrid quantum-classical topic modeling architecture that leverages quantum properties through parameterized quantum circuits. By integrating quantum-enhanced sampling into the inference pipeline, the proposed model captures richer topic distributions by mapping textual data into a higher-dimensional space. Benchmark experiments demonstrate that QTopic consistently outperforms classical approaches in terms of coherence, diversity, and topic distinctiveness, particularly when modeling a small number of topics. This study demonstrates the promise of quantum techniques in advancing unsupervised NLP, while also highlighting hardware limitations that present challenges for future research.

Collaborative Intrusion Detection System (CIDS) protect large networks against distributed attacks. However, a CIDS is vulnerable to insider attacks that decrease the mutual trust among the CIDS nodes. Most existing trust management approaches rely on a central authority, trusted third parties or network peers for managing trust. The current techniques are prone to high false positives and vulnerable to various reputation attacks. For instance, device attestation manages trust among CIDS nodes by verifying the integrity of a node's hardware and software configuration. However, it lacks real-time monitoring of the dynamic state, limiting its effectiveness against ongoing attacks and malware. Therefore, incorporating the system's dynamic state in the trust framework is crucial, but it causes false positives requiring corrective mechanisms. To address these challenges, this paper proposes a blockchain-based integrated trust management framework for CIDS, incorporating the device's genome attestation, the system's dynamic parameters, and a false positive resilient reputation mechanism. By storing the reputation scores on the blockchain, the framework alleviates the need for a third party for trust management and thus mitigates attacks applicable to reputation-based systems. The paper performs a comprehensive security and performance analysis of the proposed framework to gauge its efficiency and study the effects of a penalty on a node's reputation during the recovery and rally phases. We also study the impact of false positives on the reputation of a node. The results show that Hyperledger Fabric offers lower transaction latency and low CPU utilization compared to Ethereum Blockchain.

Phishing is one of the most prominent and long-lasting cyber-attacks, whereby attackers use social engineering methods to deceive targets to reveal private information. This study analyzes individual differences in victims’ vulnerability from the perspective of victimology and applied psychology. Although most studies have focused on the technical nature of phishing attacks, very little is known about personality traits as drivers of vulnerability. It involved a large-scale survey in which all participants completed a personality assessment questionnaire, along with a phishing susceptibility questionnaire. The results of the survey could be used to create personalized phishing prevention programs in which personality traits, which could be particularly susceptible to phishing, would be targeted. The developed treatments were evaluated in a randomized controlled trial. The findings identified crucial personality traits that influenced the tendency toward phishing attacks, specifically impulsivity and neuroticism. The designed programs for phishing prevention proved capable of reducing susceptibility, thus informing selective intervention designs for improved cybersecurity. This study underscores the importance of integrating psychological theories and victimology approaches to better understand and mitigate phishing risks, offering valuable insights for both academic and practical applications in cybersecurity.

Quantum neural networks have emerged as a promising approach to solving complex problems across various domains, especially when integrated with classical methods. Several hybrid quantum-classical architectures have been developed to leverage the potential of quantum advantages for image classification tasks. The design of the quantum layer plays an important role in exploiting quantum properties such as superposition and entanglement. In this research, we propose hybrid quantum neural networks with multiple quantum layers, utilizing sequential circuits for enhanced feature representation through structured depth, and non-sequential circuits to reduce complexity and improve performance. Our experimental results demonstrate that stacking multiple layers in the quantum circuit enhances performance significantly. Furthermore, the results indicate that the optimal range of 6–10 qubits achieves the best trade-off between accuracy and computational efficiency. The results also show that amplitude embedding consistently outperformed angle embedding for image classification tasks. Notably, our proposed hybrid sequential model with amplitude embedding outperforms traditional convolutional neural networks on MNIST and Fashion-MNIST datasets, while requiring fewer parameters. These findings provide valuable insights for advancing quantum machine learning in real-world applications.

Humans have been identified as the weakest link in the information security chain and the root cause of numerous security incidents in organisations. This has also been augmented by an increased number of employees working remotely due to the COVID-19 pandemic. In such a case, without the security protection that office systems afford and increased reliance on technology the security of the information assets of the organization now heavily relies on the employee's behaviour. While organisational culture helps to mould, guide and shape attitudes and behaviours of employees, employees' information security behaviour is a major concern. Therefore, the aim of this study was to empirically examine the relationship between organisational culture and employee's information security behaviour. An online survey was conducted over a period of two months. Organisational culture was measured using Organisational Culture Assessment Instrument (OCAI). SPSSv25 was used to do correlations, crosstabulations and regression analysis. Our main finding, was that organisational culture can significantly predict employee's online behaviour. Another observation was a negative relationship between organisational culture and security behaviour, which indicated that moving from a Clan culture towards a Hierarchy culture, the employee's security behaviours move from a naive behaviour through risk inclined to risk-averse behaviour.

Cloud computing has emerged as an efficient distribution platform in modern distributed computing offering scalability and flexibility. Task scheduling is considered as one of the main crucial aspects of cloud computing. The primary purpose of the task scheduling mechanism is to reduce the cost and makespan and determine which virtual machine (VM) needs to be selected to execute the task. It is widely acknowledged as a nondeterministic polynomial-time complete problem, necessitating the development of an efficient solution. This paper presents an innovative approach to task scheduling and allocation within cloud computing systems. Our focus lies on improving both the efficiency and cost-effectiveness of task execution, with a specific emphasis on optimizing makespan and resource utilization. This is achieved through the introduction of an Advanced Max–Min Algorithm, which builds upon traditional methodologies to significantly enhance performance metrics such as makespan, waiting time, and resource utilization. The selection of the Max–Min algorithm is rooted in its ability to strike a balance between task execution time and resource utilization, making it a suitable candidate for addressing the challenges of cloud task scheduling. Furthermore, a key contribution of this work is the integration of a cost-aware algorithm into the scheduling framework. This algorithm enables the effective management of task execution costs, ensuring alignment with user requirements while operating within the constraints of cloud service providers. The proposed method adjusts task allocation based on cost considerations dynamically. Additionally, the presented approach enhances the overall economic efficiency of cloud computing deployments. The findings demonstrate that the proposed Advanced Max–Min Algorithm outperforms the traditional Max–Min, Min–Min, and SJF algorithms with respect to makespan, waiting time, and resource utilization.

The dining sector in developing countries faces numerous challenges, including inefficiencies in order handling, resource management, and ensuring food quality and customer privacy. Traditional methods often lead to delays, errors, and dissatisfaction. This paper proposes a quick-witted, intelligent order-handling system utilizing the Internet of Things (IoT) to address these challenges and enhance the overall dining experience. We present a comprehensive approach to developing and implementing an IoT-based automated order-handling system tailored to restaurants' specific needs and challenges in developing countries, highlighting the importance of technology in enhancing operational efficiency and customer satisfaction. The proposed automated secure order-handling system using IoT demonstrates significant potential for improving efficiency and customer satisfaction in the dining sector. By addressing common problems through advanced technology, this system offers a sustainable solution that enhances the dining experience while ensuring food orders' validity, quality, and privacy. We analyzed the potential impact of implementing such a system in developing countries, focusing on economic and operational benefits.

The stock market is changing quickly, and its nonlinear characteristics make stock price prediction difficult. Predicting stock prices is challenging due to several factors, including a company’s financial performance, unforeseen circumstances, general economic conditions, politics, current assets, global situation, etc. Despite these terms, sufficient data are available to identify stock price movement trends using the different technical approaches. In this research, we empirically analyzed long short-term memory (LSTM) networks in the context of time-series prediction. Our investigation leveraged a diverse set of real-world datasets and provided quantitative insights into the performance of LSTMs. Across a spectrum of time-series forecasting tasks, LSTM models demonstrated an impressive mean absolute error (MAE) reduction of 23.4% compared to traditional forecasting methods. Specifically, LSTM achieved an average prediction accuracy of 89.7% in financial market predictions, outperforming baseline models by a significant margin. The aim is to obtain a value that can be compared to the present price of an asset to determine whether it is overvalued or undervalued, which anticipates the price patterns by analyzing previous market information, such as price and volume, compared to this stock analysis approach.

Machine learning now confronts two significant obstacles: the first is data isolation in most organizations' silos, and the second is data privacy and security enforcement. The widespread application of Machine Learning techniques in patient care is currently hampered by limited dataset availability for algorithm training and validation due to the lack of standardised electronic medical records and strict legal and ethical requirements to protect patient privacy. To avoid compromising patient privacy while supporting scientific analysis on massive datasets to improve patient care, it is necessary to analyse and implement Machine Learning solutions that fulfil data security and consumption demands. In this survey paper, we meticulously explain the existing works of federated learning from many perspectives to give a thorough overview and promote future research in this area. Then, we determine the current challenges, attack vectors and potential prospects for federated learning research. We analysed the similarities, differences and advantages between federated learning and other machine learning techniques. We also discussed about system and statistical heterogeneity and related efficient algorithms.

Social bots are automated social media accounts governed by software and controlled by humans at the backend. Some bots have good purposes, such as automatically posting information about news and even to provide help during emergencies. Nevertheless, bots have also been used for malicious purposes, such as for posting fake news or rumour spreading or manipulating political campaigns. There are existing mechanisms that allow for detection and removal of malicious bots automatically. However, the bot landscape changes as the bot creators use more sophisticated methods to avoid being detected. Therefore, new mechanisms for discerning between legitimate and bot accounts are much needed. Over the past few years, a few review studies contributed to the social media bot detection research by presenting a comprehensive survey on various detection methods including cutting-edge solutions like machine learning (ML)/deep learning (DL) techniques. This paper, to the best of our knowledge, is the first one to only highlight the DL techniques and compare the motivation/effectiveness of these techniques among themselves and over other methods, especially the traditional ML ones. We present here a refined taxonomy of the features used in DL studies and details about the associated pre-processing strategies required to make suitable training data for a DL model. We summarize the gaps addressed by the review papers that mentioned about DL/ML studies to provide future directions in this field. Overall, DL techniques turn out to be computation and time efficient techniques for social bot detection with better or compatible performance as traditional ML techniques.