Review
Computer Science, Information Systems
Shaashwat Agrawal, Sagnik Sarkar, Ons Aouedi, Gokul Yenduri, Kandaraj Piamrat, Mamoun Alazab, Sweta Bhattacharya, Praveen Kumar Reddy Maddikunta, Thippa Reddy Gadekallu
Summary: The rapid development of the Internet and smart devices has led to a surge in network traffic, making the infrastructure more complex and heterogeneous. The predominant usage of mobile phones, wearable devices, and autonomous vehicles generates a huge amount of data every day. Intrusion detection systems play a significant role in ensuring the security and privacy of these devices. Machine Learning and Deep Learning with Intrusion Detection Systems have gained momentum due to their high classification accuracy. However, the need to store and communicate data to a centralized server potentially compromises privacy and security. On the other hand, Federated Learning provides a privacy-preserving decentralized learning technique that trains models locally and transfers parameters to the centralized server. This paper aims to provide a comprehensive review of the use of Federated Learning in intrusion detection systems, discussing various types of IDS, relevant ML approaches, and associated issues. The paper also presents a detailed overview of the implementation of Federated Learning in anomaly detection and identifies the challenges and potential solutions for future research.
COMPUTER COMMUNICATIONS
(2022)
Article
Chemistry, Analytical
Shumon Alam, Yasin Alam, Suxia Cui, Cajetan Akujuobi
Summary: Cybersecurity is a critical issue in today's internet world. Classical security systems are unable to detect sophisticated zero-day attacks, making machine learning-based solutions more attractive. However, meaningful and realistic network datasets are necessary to develop an ML-based anomaly detection system. Existing public network datasets have limitations in terms of data creation process and lack of diverse attack scenarios. This work has created realistic network datasets with various attack scenarios and diverse background traffic, and compared the performance of different ML algorithms in detecting anomaly traffic. The CNN-Pseudo-AE algorithm shows promising performance compared to classical supervised algorithms.
Article
Computer Science, Interdisciplinary Applications
Justus Zipfel, Felix Verworner, Marco Fischer, Uwe Wieland, Mathias Kraus, Patrick Zschech
Summary: Visual quality assurance has transitioned from manual labor to automated assessment with the help of machine learning. However, most supervised learning approaches are limited to predefined categories, which means they can't detect new or unseen error types. This study explores unsupervised models based on deep neural networks that can assess overall object quality without category limitations. Evaluating three unsupervised models using a quality inspection case from a European car manufacturer, the results show that fully unsupervised approaches can achieve reliable performance comparable to supervised methods.
COMPUTERS & INDUSTRIAL ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Jiaxin Liu, Xucheng Song, Yingjie Zhou, Xi Peng, Yanru Zhang, Pei Liu, Dapeng Wu, Ce Zhu
Summary: With the wide deployment of edge devices, it is important and challenging to detect packet payload anomalies for the safe and efficient operations of edge applications. However, existing approaches have limitations in detecting anomalies with long-term dependency relationships and rely on in-depth expert knowledge. To overcome these limitations, a deep learning-based framework is proposed, which consists of a block sequence construction method and a detection model based on LSTM, CNN, and Multi-head Self Attention. Experimental results show that the proposed model achieves a higher detection rate and a lower false positive rate compared to traditional and state-of-the-art methods.
Article
Computer Science, Theory & Methods
Tharindu Fernando, Harshala Gammulle, Simon Denman, Sridha Sridharan, Clinton Fookes
Summary: This paper provides a thorough theoretical analysis of popular deep learning techniques in medical anomaly detection, comparing and contrasting their architectural differences as well as training algorithms. Additionally, it outlines the key limitations of existing deep medical anomaly detection techniques and proposes key research directions for further investigation.
ACM COMPUTING SURVEYS
(2021)
Article
Chemistry, Analytical
Joon-Hyung Park, Yeong-Seok Kim, Hwi Seo, Yeong-Jun Cho
Summary: Recently, companies have adopted automated defect detection methods, particularly deep learning-based image understanding techniques, in defect-free PCB manufacturing. This study analyzes the stable training of deep learning models for PCB defect detection. The characteristics of industrial images, potential factors causing image data changes, and various defect detection methods are summarized. Experimental results reveal the impact of factors such as detection methods, data quality, and image contamination on defect detection. Knowledge and guidelines for accurate PCB defect detection are presented based on the study's overview and experiment results.
Article
Computer Science, Information Systems
Irfan Ali Kandhro, Sultan M. M. Alanazi, Fayyaz Ali, Asadullah Kehar, Kanwal Fatima, Mueen Uddin, Shankar Karuppayah
Summary: Computer viruses, malicious attacks, and other hostiles can harm computer networks. Intrusion detection is crucial for network security and as an active defense technology. Traditional systems face challenges such as poor accuracy, ineffective detection, high false positives, and an inability to handle new intrusions. To address these issues, we propose a deep learning-based method to detect vulnerabilities and breaches in cyber-physical systems.
Article
Public, Environmental & Occupational Health
Colin Price, Joseph A. Russell
Summary: AMAnD is a computational tool that utilizes Deep Support Vector Data Description (DeepSVDD) models to detect anomalous metagenomic samples from typical samples. It can identify abnormal samples like COVID-19 and STC in the feature space and also detect contaminant inserts in synthetic metagenomes. The assumption-free anomaly flagging method and the real-time model training update capability of AMAnD make it suitable for a wide range of applied metagenomics biosurveillance use-cases.
FRONTIERS IN PUBLIC HEALTH
(2023)
Article
Chemistry, Analytical
Saeid Sheikhi, Panos Kostakos
Summary: In this paper, a new intrusion detection model is proposed that utilizes a genetic algorithm and optimization algorithms for training and testing on the NSL-KDD dataset. The results demonstrate improved detection capability and accuracy compared to other techniques.
Article
Chemistry, Analytical
Diulhio Candido de Oliveira, Bogdan Tomoyuki Nassu, Marco Aurelio Wehrmeister
Summary: In this paper, a one-class learning approach is introduced for detecting modifications in assembled PCBs based on photographs taken without tight control over perspective and illumination conditions. Anomaly detection and segmentation are addressed as a case of anomaly detection in an uncontrolled environment. The proposed approach outperforms other state-of-the-art approaches for anomaly segmentation in the considered scenario.
Article
Computer Science, Hardware & Architecture
Qiumei Cheng, Chunming Wu, Haifeng Zhou, Dezhang Kong, Dong Zhang, Junchi Xing, Wei Ruan
Summary: The paper proposes a novel OpenFlow-enabled deep packet inspection (OFDPI) approach for efficient and adaptive packet inspection in SDN. OFDPI conducts early detection at the flow-level granularity and deep packet inspection at the packet-level granularity, balancing detection accuracy and performance bottleneck.
JOURNAL OF NETWORK AND COMPUTER APPLICATIONS
(2021)
Article
Chemistry, Multidisciplinary
Helmut Neuschmied, Martin Winter, Branka Stojanovic, Katharina Hofer-Schmitz, Josip Bozic, Ulrike Kleb
Summary: In the face of emerging technological achievements, cyber security remains a significant issue. This paper focuses on the detection of advanced-persistent-threat (APT) attacks in network systems using machine-learning algorithms such as autoencoders. Experimental evaluation shows promising results, indicating the plausibility of features and the performance of applied algorithms. Suggestions for improvements in the anomaly detector are provided.
APPLIED SCIENCES-BASEL
(2022)
Article
Multidisciplinary Sciences
Tuan-Hong Chua, Iftekhar Salam
Summary: Cybersecurity is a major concern for organizations due to the increasing number of cyberattacks as Internet usage grows. Researchers are focusing on developing machine learning-based intrusion detection systems (IDS) to detect zero-day attacks. This study evaluates the long-term performance of ML-based IDS by using a dataset created later than the training dataset, which better reflects changes in attack types and network infrastructure over time. Experiment results show that SVM and ANN are most resistant to overfitting, while DT and RF suffer the most from overfitting despite performing well on the training dataset. All models perform well when the difference between the training and testing datasets is small, as observed in the LUFlow dataset.
Article
Energy & Fuels
Lukas Bommes, Mathis Hoffmann, Claudia Buerhop-Lutz, Tobias Pickel, Jens Hauch, Christoph Brabec, Andreas Maier, Ian Marius Peters
Summary: This study frames fault detection in PV modules as an unsupervised domain adaptation problem, training on labeled data of one source PV plant and making predictions on another target plant. By using a ResNet-34 convolutional neural network and a k-nearest neighbor classifier, the method achieves satisfactory detection results.
PROGRESS IN PHOTOVOLTAICS
(2022)
Review
Computer Science, Information Systems
Huseyin Ahmetoglu, Resul Das
Summary: The development of network technologies and the increasing amount of data transferred on networks have led to a rise in cyber threats and attacks. Machine learning offers tools and techniques for automating the detection and analysis of these attacks. This study discusses the different machine learning approaches used to detect and analyze attacks, including anomaly detection, classification, and analysis. The study also examines the performance and results of different methods, as well as the datasets used in the research.
INTERNET OF THINGS
(2022)