Array最新文献

Federated Learning (FL) has emerged as a promising approach to address privacy concerns inherent in Machine Learning (ML) practices. However, conventional FL methods, particularly those following the Centralized FL (CFL) paradigm, utilize a central server for global aggregation, which exhibits limitations such as bottleneck and single point of failure. To address these issues, the Decentralized FL (DFL) paradigm has been proposed, which removes the client–server boundary and enables all participants to engage in model training and aggregation tasks. Nevertheless, as CFL, DFL remains vulnerable to adversarial attacks, notably poisoning attacks that undermine model performance. While existing research on model robustness has predominantly focused on CFL, there is a noteworthy gap in understanding the model robustness of the DFL paradigm. In this paper, a thorough review of poisoning attacks targeting the model robustness in DFL systems, as well as their corresponding countermeasures, are presented. Additionally, a solution called DART is proposed to evaluate the robustness of DFL models, which is implemented and integrated into a DFL platform. Through extensive experiments, this paper compares the behavior of CFL and DFL under diverse poisoning attacks, pinpointing key factors affecting attack spread and effectiveness within the DFL. It also evaluates the performance of different defense mechanisms and investigates whether defense mechanisms designed for CFL are compatible with DFL. The empirical results provide insights into research challenges and suggest ways to improve the robustness of DFL models for future research.

In today’s digital and internet era, threat intelligence analysis is of paramount importance to ensure network and information security. Named Entity Recognition (NER) is a fundamental task in natural language processing, aimed at identifying and extracting specific types of named entities from text, such as person names, locations, organization names, dates, times, currencies, and more. The quality of entities determines the effectiveness of upper-layer applications such as knowledge graphs. Recently, there has been a scarcity of training data in the threat intelligence field, and single models suffer from poor generalization ability. To address this, we propose a multi-view learning model, named the Few-shot Threat Intelligence Named Entity Recognition Model (FTM). We enhance the fusion method based on FTM, and further propose the FTM-GRU (Gate Recurrent Unit) model. The FTM model is based on the Tri-training algorithm to collaboratively train three few-shot NER models, leveraging the complementary nature of different model views to enable them to capture more threat intelligence domain knowledge at the coding level.FTM-GRU improves the fusion of multiple views. FTM-GRU uses the improved GRU model structure to control the memory and forgetting of view information, and introduces a relevance calculation unit to avoid redundancy of view information while highlighting important semantic features. We label and construct a few-shot Threat Intelligence Dataset (TID), and experiments on TID as well as the publicly available National Vulnerability Database (NVD) validate the effectiveness of our model for NER in the threat intelligence domain. Experimental results demonstrate that our proposed model achieves better recognition results in the task.

The increasing use of unmanned aerial vehicles (UAVs) in both military and civilian applications, such as infrastructure inspection, package delivery, and recreational activities, underscores the importance of enhancing their autonomous functionalities. Artificial intelligence (AI), particularly deep learning-based computer vision (DL-based CV), plays a crucial role in this enhancement. This paper aims to provide a systematic literature review (SLR) of Scopus-indexed research studies published from 2019 to 2024, focusing on DL-based CV approaches for autonomous UAV applications. By analyzing 173 studies, we categorize the research into four domains: sensing and inspection, landing, surveillance and tracking, and search and rescue. Our review reveals a significant increase in research utilizing computer vision for UAV applications, with over 39.5 % of studies employing the You Only Look Once (YOLO) framework. We discuss the key findings, including the dominant trends, challenges, and opportunities in the field, and highlight emerging technologies such as in-sensor computing. This review provides valuable insights into the current state and future directions of DL-based CV for autonomous UAVs, emphasizing its growing significance as legislative frameworks evolve to support these technologies.

The data model is a critical component of an Adaptive Web System (AWS). The major goals of such a data model are describing the application domain of the AWS and capturing data about the user in order to support the “adaptation effect”. There have been many proposals for data models, principally based on knowledge representation, machine learning, logic and reasoning, and, recently, ontologies. These models are focused on the implementation of the core layer of AWS, that is realizing the adaptation of contents and presentations of the system, but sometimes they are poor with respect to the application domain design. In this paper, we present an extension of the state-of-the-art XML Adaptive Hypermedia Model (XAHM), Object-Oriented XAHM (OO-XAHM) that supports the application domain modeling using an object-oriented approach. We also provide the formal definition of the model, its description via Unified Modeling Language (UML), and its implementation using XML Schema. Finally, we provide a complete case study that focuses the attention on the well-known Italian archaeological site Pompeii.

Otitis media (OM) is a common infection or inflammation of the middle ear causing conductive hearing loss that primarily affects children and may delay speech, language, and cognitive development. OM can manifest itself in different forms, and can be diagnosed using (video) otoscopy (visualizing the tympanic membrane) or (video) pneumatic otoscopy and tympanometry. Accurate diagnosis of OM is challenging due to subtle differences in otoscopic features. This research aims to develop an automated computer-aided design (CAD) system to assist clinicians in diagnosing OM using otoscopy images. The ground truths, generated manually and validated by otolaryngologists, are utilized to train the proposed nested U-Net++ model. Ten clinically relevant gray level co-occurrence matrix (GLCM) and morphological features were extracted from the segmented Region of Interest (ROI) and validated for OM classification based on a statistical significance test. These features serve as input for a Graph Neural Network (GNN) model, the base model in our research. An optimized GNN model is proposed after ablation study of the base model. Three datasets, one private dataset, and two public ones have been used, where the private dataset is utilized for both training and testing, and the public datasets are used to test the robustness of the proposed GNN model only. The proposed GNN model obtained the highest accuracy in diagnosing OM: 99.38 %, 93.51 %, and 91.38 % for the private dataset, public dataset1, and public dataset2, respectively. The proposed methodology and results of this research might enhance clinicians' effectiveness in diagnosing OM.

Glaucoma is a leading cause of irreversible visual field degradation, significantly impacting ocular health. Timely identification and diagnosis of this condition are critical to prevent vision loss. A range of diagnostic techniques is employed to achieve this, from traditional methods reliant on expert interpretation to modern, fully computerized diagnostic approaches. The integration of computerized systems designed for the early detection and classification of clinical indicators of glaucoma holds immense potential to enhance the accuracy of disease diagnosis. Pupillary Light Reflex (PLR) analysis emerges as a promising avenue for glaucoma screening, mainly due to its cost-effectiveness compared to exams such as Optical Coherence Tomography (OCT), Humphrey Field Analyzer (HFA), and fundoscopic examinations. The noninvasive nature of PLR testing obviates the need for disposable components and agents for pupil dilation. This facilitates multiple successive administrations of the test and enables the possibility of remote execution. This study aimed to improve the automated diagnosis of glaucoma using PLR data, conducting an extensive comparative analysis incorporating neural networks and machine learning techniques. It also compared the performance of different data processing methods, including filtering techniques, feature extraction, data balancing, feature selection, and their effects on classification. The findings offer insights and guidelines for future methodologies in glaucoma screening utilizing pupillary light response signals.

This paper explores the challenges and potential solutions related to data collection, integration, processing, and utilization in defense manufacturing within the context of Industry 4.0. While Industry 4.0 envisions the integration of various technologies to achieve seamless operations in industries, the unique characteristics of defense manufacturing, such as stringent data limitations and security requirements, make direct translation challenging. Through a comprehensive review of academic literature, key themes were identified, including quality control, digitalization, cyber–physical aspects, sustainability, risk management, ownership of information, and security. Drawing from the reviewed publications, potential solutions were distilled into related approaches, such as data governance frameworks, data exchange standards, blockchain, additive manufacturing, transparent digital supply chains, and smart factories. These solutions present opportunities for the Australian defense manufacturing industry to overcome the identified challenges and leverage the benefits of Industry 4.0, including improved quality control, increased efficiency, enhanced security, and optimized supply chains. By embracing these opportunities, the defense manufacturing sector can successfully navigate the complexities of Industry 4.0 and realize its vision of seamless integration for continued growth and success.

Over the past two decades, computer-aided detection and diagnosis have emerged as a field of research. The primary goal is to enhance the diagnostic and treatment procedures for radiologists and clinicians in medical image analysis. With the help of big data and advanced artificial intelligence (AI) technologies, such as machine learning and deep learning algorithms, the healthcare system can be made more convenient, active, efficient, and personalized. The primary goal of this literature survey was to present a thorough overview of the most important developments related to computer-aided diagnosis (CAD) systems in medical imaging. This survey is of considerable importance to researchers and professionals in both medical and computer sciences. Several reviews on the specific facets of CAD in medical imaging have been published.

Nevertheless, the main emphasis of this study was to cover the complete range of capabilities of CAD systems in medical imaging. This review article introduces background concepts used in typical CAD systems in medical imaging by outlining and comparing several methods frequently employed in recent studies. This article also presents a comprehensive and well-structured survey of CAD in medicine, drawing on a meticulous selection of relevant publications. Moreover, it describes the process of handling medical images and introduces state-of-the-art AI-based CAD technologies in medical imaging, along with future directions of CAD. This study indicates that deep learning algorithms are the most effective method to diagnose and detect diseases.

To protect systems from malicious activities, it is important to differentiate between valid and harmful commands. One way to achieve this is by learning the syntax of the commands, which is a complex task because of the expansive and evolving nature of command syntax. To address this, we harnessed the power of a language model. Our methodology involved constructing a specialized vocabulary from our commands dataset, and training a custom tokenizer with a Masked Language Model head, resulting in the development of a BERT-like language model. This model exhibits proficiency in learning command syntax by predicting masked tokens. In comparative analyses, our language model outperformed the Markov Model in categorizing commands using clustering algorithms (DBSCAN, HDBSCAN, OPTICS). The language model achieved higher Silhouette scores (0.72, 0.88, 0.85) compared to the Markov Model (0.53, 0.25, 0.06) and demonstrated significantly lower noise levels (2.63%, 5.39%, 8.49%) versus the Markov Model’s higher noise rates (9.31%, 29.85%, 50.35%). Further validation with manually crafted syntax and BERTScore assessments consistently produced metrics above 0.90 for precision, recall, and F1-score. Our language model excels at learning command syntax, enhancing protective measures against malicious activities.

Privacy Impact Assessments (PIAs) offer a process for assessing the privacy impacts of a project or system. As a privacy engineering strategy, they are one of the main approaches to privacy by design, supporting the early identification of threats and controls. However, there is still a shortage of empirical evidence on their use and proven effectiveness in practice. To better understand the current literature and research, this paper provides a comprehensive Scoping Review (ScR) on the topic of PIAs “in the wild,” following the well-established Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines. This ScR includes 45 studies, providing an extensive synthesis of the existing body of knowledge, classifying types of research and publications, appraising the methodological quality of primary research, and summarising the positive and negative aspects of PIAs in practice, as reported by those studies. This ScR also identifies significant research gaps (e.g., evidence gaps from contradictory results and methodological gaps from research design deficiencies), future research pathways, and implications for researchers, practitioners, and policymakers developing and using PIA frameworks. As we conclude, there is still a significant need for more primary research on the topic, both qualitative and quantitative. A critical appraisal of qualitative studies revealed deficiencies in the methodological quality, and only four quantitative studies were identified, suggesting that current primary research remains incipient. Nonetheless, PIAs can be regarded as a prominent sub-area in the broader field of empirical privacy engineering, in which further scientific research to support existing practices is needed.