Search

This study aims to perform a thorough systematic review investigating and synthesizing existing research on defense strategies and methodologies in adversarial attacks using machine learning (ML) and deep learning methods. A methodology was conducted to guarantee a thorough literature analysis of the studies using sources such as ScienceDirect, Scopus, IEEE Xplore, and Web of Science. A question was shaped to retrieve articles published from 2019 to April 2024, which ultimately produced a total of 704 papers. A rigorous screening, deduplication, and matching of the inclusion and exclusion criteria were followed, and hence 42 studies were included in the quantitative synthesis. The considered papers were categorized into a coherent and systematic classification including three categories: security enhancement techniques, adversarial attack strategies and defense mechanisms, and innovative security mechanisms and solutions. In this article, we have presented a systematic and comprehensive analysis of earlier studies and opened the door to potential future studies by discussing in depth four challenges and motivations of adversarial attacks, while three recommendations have been discussed. A systematic science mapping analysis was also performed to reorganize and summarize the results of studies to address the issues of trustworthiness. Moreover, this research covers a large variety of network and cybersecurity applications of defense in adversarial attack subjects, including intrusion detection systems, anomaly detection, ML-based defenses, and cryptographic techniques. The relevant conclusions well demonstrate what have achieved in defense mechanisms against adversarial attacks. In addition, the analysis revealed a few emerging tendencies and deficiencies in the area to be remedied through better and more dependable mitigation methods against advanced persistent threats. The findings of this review have crucial implications for the community of researchers, practitioners, and policy makers in network and cybersecurity using artificial intelligence applications.

Abstract This study delves into the complex prioritization process for Autism Spectrum Disorder (ASD), focusing on triaged patients at three urgency levels. Establishing a dynamic prioritization solution is challenging for resolving conflicts or trade-offs among ASD criteria. This research employs fuzzy multi-criteria decision making (MCDM) theory across four methodological phases. In the first phase, the study identifies a triaged ASD dataset, considering 19 critical medical and sociodemographic criteria for the three ASD levels. The second phase introduces a new Decision Matrix (DM) designed to manage the prioritization process effectively. The third phase focuses on the new extension of Fuzzy-Weighted Zero-Inconsistency (FWZIC) to construct the criteria weights using Single-Valued Neutrosophic 2-tuple Linguistic (SVN2TL). The fourth phase formulates the Multi-Attributive Border Approximation Area Comparison (MABAC) method to rank patients within each urgency level. Results from the SVN2TL-FWZIC weights offer significant insights, including the higher criteria values "C12 = Laughing for no reason" and "C16 = Notice the sound of the bell" with 0.097358 and 0.083832, indicating their significance in identifying potential ASD symptoms. The SVN2TL-FWZIC weights offer the base for prioritizing the three triage levels using MABAC, encompassing medical and behavioral dimensions. The methodology undergoes rigorous evaluation through sensitivity analysis scenarios, confirming the consistency of the prioritization results with critical analysis points. The methodology compares with three benchmark studies, using four distinct points, and achieves a remarkable 100% congruence with these prior investigations. The implications of this study are far-reaching, offering a valuable guide for clinical psychologists in prioritizing complex cases of ASD patients.

Abstract In the context of autism spectrum disorder (ASD) triage, the robustness of machine learning (ML) models is a paramount concern. Ensuring the robustness of ML models faces issues such as model selection, criterion importance, trade-offs, and conflicts in the evaluation and benchmarking of ML models. Furthermore, the development of ML models must contend with two real-time scenarios: normal tests and adversarial attack cases. This study addresses this challenge by integrating three key phases that bridge the domains of machine learning and fuzzy multicriteria decision-making (MCDM). First, the utilized dataset comprises authentic information, encompassing 19 medical and sociodemographic features from 1296 autistic patients who received autism diagnoses via the intelligent triage method. These patients were categorized into one of three triage labels: urgent, moderate, or minor. We employ principal component analysis (PCA) and two algorithms to fuse a large number of dataset features. Second, this fused dataset forms the basis for rigorously testing eight ML models, considering normal and adversarial attack scenarios, and evaluating classifier performance using nine metrics. The third phase developed a robust decision-making framework that encompasses the creation of a decision matrix (DM) and the development of the 2-tuple linguistic Fermatean fuzzy decision by opinion score method (2TLFFDOSM) for benchmarking multiple-ML models from normal and adversarial perspectives, accomplished through individual and external group aggregation of ranks. Our findings highlight the effectiveness of PCA algorithms, yielding 12 principal components with acceptable variance. In the external ranking, logistic regression (LR) emerged as the top-performing ML model in terms of the 2TLFFDOSM score (1.3370). A comparative analysis with five benchmark studies demonstrated the superior performance of our framework across all six checklist comparison points.

Large Language Models (LLMs) have become a hot topic in AI due to their ability to mimic human conversation. This study compares the open artificial intelligence generative pretrained transformer-4 (GPT-4) model, based on the (GPT), and Google's artificial intelligence (AI), which is based on the Bidirectional Encoder Representations from Transformers (BERT) framework in terms of the defined capabilities and the built-in architecture. Both LLMs are prominent in AI applications. First, eight different capabilities were identified to evaluate these models, i.e. translation accuracy, text generation, factuality, creativity, intellect, deception avoidance, sentiment classification, and sarcasm detection. Next, each capability was assessed using instructions. Additionally, a categorized LLM evaluation system has been developed by means of using ten research questions per category based on this paper's main contributions from a prompt engineering perspective. It should be highlighted that GPT-4 and Google AI successfully answered 85 % and 68,7 % of the study prompts, respectively. It has been noted that GPT-4 better understands prompts than Google AI, even with verbal flaws, and tolerates grammatical errors. Moreover, the GPT-4 based approach was more precise, accurate, and succinct than Google AI, which was sometimes verbose and less realistic. While GPT-4 beats Google AI in terms of translation accuracy, text generation, factuality, intellectuality, creativity, and deception avoidance, Google AI outperforms the former when considering sarcasm detection. Both sentiment classification models did work properly. More importantly, a human panel of judges was used to assess and evaluate the model comparisons. Statistical analysis of the judges' ratings revealed more robust results based on examining the specific uses, limitations, and expectations of both GPT-4 and Google AI-based approaches. Finally, the two approaches' transformers, parameter sizes, and attention mechanisms have been examined.

Adversarial attacks pose a significant threat to deep learning models, specifically medical images, as they can mislead models into making inaccurate predictions by introducing subtle distortions to the input data that are often imperceptible to humans. Although adversarial training is a common technique used to mitigate these attacks on medical images, it lacks the flexibility to address new attack methods and effectively improve feature representation. This paper introduces a novel Model Ensemble Feature Fusion (MEFF) designed to combat adversarial attacks in medical image applications. The proposed model employs feature fusion by combining features extracted from different DL models and then trains Machine Learning classifiers using the fused features. It uses a concatenation method to merge the extracted features, forming a more comprehensive representation and enhancing the model's ability to classify classes accurately. Our experimental study has performed a comprehensive evaluation of MEFF, considering several challenging scenarios, including 2D and 3D images, greyscale and colour images, binary classification, and multi-label classification. The reported results demonstrate the robustness of using MEFF against different types of adversarial attacks across six distinct medical image applications. A key advantage of MEFF is its capability to incorporate a wide range of adversarial attacks without the need to train from scratch. Therefore, it contributes to developing a more diverse and robust defence strategy. More importantly, by leveraging feature fusion and ensemble modelling, MEFF enhances the resilience of DL models in the face of adversarial attacks, paving the way for improved robustness and reliability in medical image analysis.

In recent years, calls have increased for adherence to standards that ensure sustainability, including the global initiative presented by the United Nations with 17 Sustainable Development Goals (SDGs) to ensure a more sustainable future. Achieving these goals is extremely important, as institutions have sought to integrate technology, especially business intelligence, into their operations to ensure their achievement. This study aims to provide a systematic literature review of the intersection of information systems and sustainability in business intelligence. Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology was utilized to select high-quality studies from various databases, including ScienceDirect, IEEE Xplore, and Scopus, to be included in this review. The methodology resulted in 32 studies taxonomized into four main categories covering different aspects of the intersection of information systems and sustainability. This study discusses integrating information systems and sustainability in various sectors, such as tourism, health, urban, and other sectors, with different technologies, such as Blockchain, IoT, Industry 4.0, and other innovations. Moreover, the information system types implemented to support sustainability practices in different domains are highlighted.