Journal of Artificial Intelligence, Applications and Innovations

Federated Learning is a distributed machine learning paradigm designed to preserve user privacy on decentralized devices without transferring raw data to a central server. Protecting data privacy in FL involves determining permissible operations and how they can be executed. This review provides an in-depth exploration of privacy threat models within FL, distinguishing between scenarios where the central server is either trusted or untrusted, and identifying appropriate defensive tools and technologies for these settings. The review covers secure computational techniques, including MPC, HE, and TEEs, as well as privacy-preserving mechanisms such as DP, LDP, and DDP models. It also examines hybrid approaches that combine multiple privacy models to enhance efficiency and robustness. The effectiveness of these methods is analysed across different scenarios involving both honest and potentially malicious servers and users. The findings reveal that while privacy-preserving methods mitigate risks, challenges persist in trade off privacy, communication efficiency, and model accuracy. This review highlights open research directions and serves as a comprehensive reference for researchers and practitioners seeking to implement robust privacy measures in federated learning systems.

Security anomaly detection, a critical element in safeguarding digital systems, has undergone a transformative evolution through the integration of deep learning techniques. This comprehensive review navigates the landscape of security anomaly detection, unveiling the potential and challenges within this realm. The research methodology involved systematic data collection from renowned databases, including Scopus, Web of Science, and Google Scholar. Key topics explored include the integration of deep learning models, benchmark datasets, preprocessing techniques, ethical considerations, and future directions. Deep learning models, such as autoencoders, recurrent neural networks (RNNs), and convolutional neural networks (CNNs), have proven invaluable in enhancing detection accuracy and efficiency. Benchmark datasets like NSL-KDD, CICIDS2017, and UNSW-NB15 have emerged as essential evaluation tools. Tailored preprocessing techniques ensure data readiness for these models. Challenges encompass data imbalance, model interpretability, adversarial attacks, and scalability. Ethical and privacy considerations emphasize privacy preservation, fairness, transparency, and accountability. The convergence of deep learning with security anomaly detection heralds a new era in cybersecurity. While challenges persist, a commitment to ethical principles and exploration of innovative avenues are set to realize the full potential of deep learning for robust, efficient, and responsible security anomaly detection systems, ensuring a safer digital landscape for all.

Today, smartphones are prevalent for personal and corporate use and have become the new personal computer due to their portability, ease of use, and functionality (such as video conferencing, Internet browsing, e-mail, continuous wireless and data connectivity, worldwide map location services, and countless mobile applications such as banking applications). On the other hand, we store many sensitive and private information daily on smart devices. This information is of interest to malicious writers who are developing malware to steal information from mobile devices. Unfortunately, the open source and widespread adoption of the Android operating system has made it the most targeted of the four popular mobile platforms by malware writers. Many researchers have tried to identify malware using program signatures, which have been successful to some extent. However, the signature cannot effectively identify new and unknown malware. For this reason, in this article, we propose a method that designs a machine-learning model for Android malware detection based on the properties of Permissions, Intents APKs. In this study, we evaluated more than 25,000 Android samples belonging to malware and trusted samples. Experimental results show the effectiveness of the proposed method by obtaining 96.27% accuracy.

Word Sense Disambiguation (WSD) involves assigning the appropriate sense to ambiguous words. WSD is one of the most challenging problems in several Natural Language Processing (NLP) tasks, such as machine translation. This paper proposes a novel approach consisting of four main components. In the first part, a mining process is used to construct a tree structure that represents helpful knowledge about the conceptual relationships between each ambiguous word and its relevant context. In the second part, a Knowledge Base (KB) is constructed based on the chains derived from the tree structure. The third part involves designing an expert system for lexical ambiguity resolution using the forward chaining strategy. In the final part, the KB is upgraded to improve its effectiveness in determining the correct senses of ambiguous words. The performance of the proposed approach is evaluated on the TWA corpus. The results demonstrate the effectiveness of the proposed expert system.

Efficient partitioning of the atomic space among parallel FPGAs is crucial for accelerating molecular simulations. Existing research has primarily focused on uniform partitioning, assuming a homogeneous distribution of atoms. However, in scenarios with non-uniform atomic distributions, these approaches may lead to suboptimal performance. This study investigates the impact of non-uniform atom distributions on molecular simulation performance across parallel FPGAs. We propose a novel space partitioning scheme that optimizes the distribution of atomic space among FPGAs, taking into account the spatial heterogeneity of atoms. Our evaluation demonstrates that the proposed scheme consistently outperforms uniform partitioning in terms of simulation speed across various spatial dimensions and atom counts, particularly in scenarios with non-uniform atom distributions.

Cardiovascular Diseases (CVD) are significant global cause of mortality. This paper focuses on early detection of a specific type of CVD, Atrial Fibrillation (AF), through a simple approach. The methodology is based on efficient risk assessment methods to identify high-risk individuals with a comparative analysis of seven ML algorithms to find the simplest and most effective approach. The research utilizes the Sleep Heart Health Study (SHHS) dataset, a large-scale cohort study with diverse clinical parameters and polysomnographic data, which seems to be ideal for early AF prediction. The study formulates a predictive analysis based on minimal accessible data (i.e. no signal, image, or complex measurement are considered) and evaluates seven ML algorithms including Support Vector Machines (SVM), K-Nearest Neighbor (KNN), Random Forest (RF), Decision Tree (DT), Gradient Boosting (GB), Multi-Layer Perceptron (MLP), and Logistic Regression (LR). Among these methods, LR shows notable predictive accuracy. The analysis covers a diverse cohort, including various races (i.e. White, Black, and others), ethnicities, and both genders, with a focus on individuals with aged averagely more than 63. The study concludes that our formulation with the simple and readily accessible parameters predict AF reasonably well, potentially enabling early interventions to reduce morbidity and mortality.

In contemporary times, a substantial number of financial transactions and monetary transfers take place on the Internet and within electronic environments, thereby incentivizing fraudsters to infiltrate this domain. Consequently, the identification of individuals' identities in electronic service provision is exceedingly vital and crucial. This article aims to fraud detection in the banking system and present an optimal method utilizing artificial intelligence tools and model evaluation on the bank information of the Development and Cooperation Cooperative. In the initial phase, a gradient boosting algorithm, chosen for its high computational speed, is employed to train on a set of input data to identify and classify patterns of suspicious behaviors. In the second phase, an algorithm based on gradient boosting is utilized to refine results and optimize accuracy. To evaluate this approach, real data from a bank is employed, and the obtained results demonstrate that this method significantly enhances the speed and accuracy of fraud detection.

Federated learning is a novel way of training machine learning models on data that is distributed across multiple devices, such as smartphones and IoT sensors, without compromising privacy, efficiency, or security. However, federated learning faces a significant challenge when the data on each device is not independent and identically distributed (non-IID), which means that the data may have different distributions, sizes, or qualities. non-IID data is a major challenge for federated learning, as it affects the accuracy and participation of the local devices. Most existing methods focus on improving the model, algorithm, or framework of federated learning to deal with non-IID data. However, there is a lack of systematic and up-to-date reviews on this topic. In this paper, we survey different approaches to address the challenge of non-IID data in Vertical Federated Learning (VFL) and Horizontal Federated Learning (HFL). We organize the existing literature based on the perspective of the researcher and the sub-tasks involved in each approach. Our goal is to provide a comprehensive and systematic overview of the problem and its solutions.