Emerging Research on Neural Networks and Anomaly Detection

Dear Colleagues,

In this Special Issue, we aim to present the latest research trends in neural networks and anomaly detection. Particularly, we encourage the innovative application of neural networks to anomaly detection in the real world, which has been an active research area over the past few decades. Traditional machine learning-based or statistical solutions have been designed to achieve anomaly detection. However, it is challenging to apply traditional solutions to solve complex problems in various scenarios since these solutions require explicit feature extraction that typically fails to learn implicit relationships among features in the latent space. This issue has become a bottleneck to the improvement of the performance of traditional solutions when applied to anomaly detection.

Neural networks, also known as artificial neural networks or simulated neural networks, have been proposed as a promising solution for the detection of anomalies. In many cases, thanks to the capability of modeling and learning of the latent feature space, neural networks achieve a significantly better performance than that of the aforementioned traditional solutions. More specifically, solutions based on neural networks such as convolutional neural network, recurrent neural network, and autoencoder neural network have been leveraged to detect anomalies among various types of inputs, such as image, audio, and video. Large language model-based solutions are one of the emerging directions that combine advanced techniques (e.g., embedding representations, attention mechanism) to address practically challenging problems that remain in the real world.

Through this Special Issue, we hope to provide a collection of emerging research into neural networks that inspires researchers in both academia and industry to address challenges in anomaly detection. We welcome research studies on relevant topics including (but not limited to) network security, system security, mobile platforms, explainable AI, and privacy, among others.

Dr. Jiaping Gui
Dr. Futai Zou
Guest Editors

Manuscript Submission Information

Manuscripts should be submitted online at www.mdpi.com by registering and logging in to this website. Once you are registered, click here to go to the submission form. Manuscripts can be submitted until the deadline. All submissions that pass pre-check are peer-reviewed. Accepted papers will be published continuously in the journal (as soon as accepted) and will be listed together on the special issue website. Research articles, review articles as well as short communications are invited. For planned papers, a title and short abstract (about 100 words) can be sent to the Editorial Office for announcement on this website.

Submitted manuscripts should not have been published previously, nor be under consideration for publication elsewhere (except conference proceedings papers). All manuscripts are thoroughly refereed through a single-blind peer-review process. A guide for authors and other relevant information for submission of manuscripts is available on the Instructions for Authors page. Information is an international peer-reviewed open access monthly journal published by MDPI.

Please visit the Instructions for Authors page before submitting a manuscript. The Article Processing Charge (APC) for publication in this open access journal is 1600 CHF (Swiss Francs). Submitted papers should be well formatted and use good English. Authors may use MDPI's English editing service prior to publication or during author revisions.

• neural networks
• supervised learning
• unsupervised learning
• anomaly detection
• outlier detection
• machine learning

Title: Optimize Face Detection by Incorporating Classification Networks to Minimize False Positive Rates
Authors: Hui Cui
Affiliation: Monash University, Melbourne, Australia
Abstract: The rise of convolutional neural networks (CNNs) has significantly progressed face detection, notably improving accuracy and recall metrics. Precision and recall are crucial in assessing CNN-based detection models, yet a common tendency is to prioritize true positive rates while over looking false positives. A key factor contributing to this imbalance is the absence of pseudo-face images in training and evaluation datasets. This shortfall undermines the regression capabilities of detection models, resulting in numerous incorrect detections and subpar localization. To address this gap, we introduce the WIDERFACE dataset, which incorporates a substantial proportion of pseudo-face images synthesized by merging human and animal facial features. This dataset is specifically designed to enhance false positive detection in training scenarios. Additionally, we present a novel face detection framework that integrates a classification model into the existing face detection model to reduce the false positive rate and improve detection accuracy. Comparative evaluations on the WIDERFACE and other popular datasets demonstrate that our framework achieves a reduced false positive rate while maintaining the true positive rate relative to existing leading face detection models.

Title: Multi-identity Recognition of Darknet Vendors Based on Metric Learning
Authors: Yilei WANG; Xin SUN; Jiajia HAN; Yuelin Hu; Wenliang Xu; Futai Zou
Affiliation: School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University, Shanghai, China
Abstract: Dark web vendor identification can be seen as an authorship aliasing problem, aiming to determine whether different accounts on different markets belong to the same real-world vendor, in order to locate cybercriminals involved in dark web market transactions. Existing open-source datasets for dark web marketplaces are outdated and cannot simulate real-world situations, while data labeling methods are difficult and suffer from issues such as inaccurate labeling and limited cross-market research. The problem of identifying vendors’ multiple identities on the dark web involves a large number of categories and a limited number of samples, making it difficult to use traditional multiclass classification models. To address these issues, this paper proposes a metric learning-based method for dark web vendor identification, collecting product data from 21 currently active English dark web marketplaces and using a multi-dimensional feature extraction method based on product titles, descriptions, and images. Using pseudo-labeling technology combined with manual labeling improves data labeling accuracy compared to previous labeling methods. The proposed method uses a Siamese neural network with metric learning to learn the similarity between vendors and achieve the recognition of vendors’ multiple identities. This method achieved better performance with an average F1-Score of 0.889 and an accuracy rate of 97.54% on the constructed dataset. The contributions of this paper lie in the proposed method for collecting and labeling data for dark web marketplaces and overcoming the limitations of traditional multiclass classifiers to achieve effective recognition of vendors’ multiple identities.

Title: Painting Surface Defect Inspection Using Deeph Learning Model with Ship Painting Quality Evaluation Dataset
Authors: Haneol Seo; Chan-Su Lee
Affiliation: Yeungnam Univesity
Abstract: This paper p resents a painting surface defect detection model using anchor-free detection deep learning models. The dataset for ship painting quality evaluation, which provides the defect type and location, is used for the training and evaluation of the model. We used YOLOX-S and YOLOX-L models for surface defect detection from the selected defects of the dataset. The experiment results show that our proposed model achieves mAP@50:95: 0.678 using the YOLOX-L model, which outperformed the baseline model. We also compare inference time in YOLOX-S and YOLOX-L models. The proposed deep learning model shows a potential solution for paint defect detection in automotive industries.

Title: Fuzzy-based Routing Protocol to Reduce Overhead in Vehicular Opportunistic Networks
Authors: Ermioni Qafzezi; Kevin Bylykbashi; Shunya Higashi; Phudit Ampririt; Keita Matsuo; Leonard Barolli
Affiliation: Department of Information and Communication Engineering, Fukuoka Institute of Technology (FIT)
Abstract: Opportunistic Networks, characterized by intermittent connectivity and dynamic topologies, pose significant challenges for efficient message delivery, resource management, and routing decision-making. This paper introduces the Fuzzy Control Routing Protocol (FCRP), a novel approach designed to address these challenges by leveraging fuzzy logic to enhance routing decisions and improve overall network performance. FCRP evaluates key parameters, including buffer occupancy, angle to the destination, and the number of unique connections, to make context aware routing decisions. The protocol was implemented and evaluated using the FuzzyC simulator for theoretical validation and The ONE Simulator for realistic network scenarios. Simulation results demonstrate that FCRP achieves competitive delivery probability, efficient resource utilization, and low overhead compared to Epidemic and MaxProp protocols. Notably, FCRP consistently outperformed the Epidemic protocol across all metrics and exhibited comparable delivery probability to MaxProp while maintaining significantly lower overhead, particularly in high-density scenarios. These results underscore FCRP’s ability to adapt to varying network conditions, effectively balance forwarding and resource management, and maintain robust performance in dynamic vehicular environments. This study highlights FCRP’s potential as an efficient routing solution for Opportunistic Networks, particularly in applications requiring reliable communication and resource efficiency. Future work will explore further optimization to reduce latency and extend FCRP’s applicability to more complex networking environments.