ABSTRACT Research on multibiometric authentication has been actively conducted to defend against adversarial attacks targeting specific biometric modalities. Multibiometric authentication systems select backbone models and fusion methods based on biometric types to extract patterns and context, and then integrate them as probability or feature values for user authentication. However, existing systems often select these components based solely on authentication performance, without evaluating vulnerability to adversarial attacks, which reduces system reliability. This paper proposes a vulnerability analysis system for adversarially robust multibiometric authentication, analysing the impact of backbone and fusion configurations. The system consists of a modelling stage, in which backbone models (VGGNet, ResNet, ViT, BEiT) and fusion methods (general, hierarchical, dense fusion) are selected, and an adversarial attack stage, in which adversarial noise is injected using FGSM and PGD. Experimental results show that for fingerprints, the ViT backbone with general fusion is most robust due to global pattern extraction and local context preservation, while for faces, the VGGNet backbone with dense fusion is most robust by extracting local patterns and preserving both local and global context. However, for adversarial attacks on faces, as noise intensity increases, the ViT backbone capable of global pattern extraction is more robust than the VGGNet backbone. In addition, the BEiT backbone, pretrained via masked image modelling that reconstructs object structure, is vulnerable to adversarial noise. Finally, the weighted mean method, which integrates extracted patterns and context as probability values for individual analysis, is more robust than concatenation, which integrates them as feature values for joint analysis. Fingerprints, with strong linear features, are also more robust than faces, which emphasise local object characteristics.

Abstract Research on multi‐biometric authentication systems using multiple biometric modalities to defend against adversarial attacks is actively being pursued. These systems authenticate users by combining two or more biometric modalities using score or feature‐level fusion. However, research on adversarial attacks and defences against each biometric modality within these authentication systems has not been actively conducted. In this study, we constructed a multi‐biometric authentication system using fingerprint, palmprint, and iris information from CASIA‐BIT by employing score and feature‐level fusion. We verified the system's vulnerability by deploying adversarial attacks on single and multiple biometric modalities based on the FGSM, with epsilon values ranging from 0 to 0.5. The experimental results show that when the epsilon value is 0.5, the accuracy of the multi‐biometric authentication system against adversarial attacks on the palmprint and iris information decreases from 0.995 to 0.018 and 0.003, respectively, and the f1‐score decreases from 0.995 to 0.007 and 0.000, respectively, demonstrating susceptibility to adversarial attacks. In the case of fingerprint data, however, the accuracy and f1‐score decreased from 0.995 to 0.731 and from 0.995 to 0.741, respectively, indicating resilience against adversarial attacks.

Abstract Implementing the necessary countermeasures to detect the growing and highly destructive family of malware is an urgent obligation. The proliferation and diversity of malware make these problems more challenging. For beginners, it is arduous to attain crucial features for multi‐class family classification and extract valuable information from the obtained features. Another issue is that building a classification model that effectively absorbs multi‐class samples and adapts to various features is challenging. This work indicates a precise identification method for Android application families (ANDF) to tackle these issues. It perceptively analyzes the features that multi‐class families can utilize to identify members and further excavates the relationship between implicit information and the severity of those distinctions. A more appropriate classification model is developed for the heterogeneous file formats, and a more beneficial feature with a diverse array of heterogeneous information is chosen as the replacement representation of the sample. It is capable of upgrading learning ability and mastering the multi‐modal traits of the family malware. The application of ANDF to real data sets yields effective classification results. It is capable of 0.9800 in f1‐macro and has a classification accuracy of 98.61%. It performs, respectively, 0.0088 points better than the two‐feature comparison classification model and 0.0872 points better than the single‐feature comparison classification model. The kappa coefficient can also exceed 0.9830, which is at least 0.1044 higher than other contrasting classifiers and is 0.0105 greater than that of the contrasted model containing two features, which is 0.1046 larger than the classifier with a contrasting single feature.

Abstract Single image super‐resolution (SISR) has gained significant attention in image processing and computer vision, driven by deep learning‐based models like convolutional neural networks (CNN). Yet, the resource‐intensive nature of these models poses challenges when deploying them on edge devices. To address this issue, resource‐constrained models need to be developed. While recent models like the information distillation network (IDN), the information multi‐distillation network (IMDN), the residual feature distillation network (RFDN), and so on, have attempted to reduce parameters and computational complexity, further optimization remains vital. Therefore, this paper presents an approach to enhancing the efficiency and lightweight nature of the SISR. We introduce a novel lightweight SR model by building upon the RFDN architecture, the winner of the AIM2020 and NTIRE2022 SR challenges. The proposed depthwise channel attention network (DWCAN) model makes some key changes to RFDN. First, it replaces the main residual feature distillation block (RFDB) with a depthwise channel attention block (DWCAB). Additionally, DWCAN includes a shallow residual block (SRB) with depthwise separable convolution (DW) and a channel attention (CA) block. The primary goal of our work is to significantly reduce model parameters, computational operations, inference time, and memory size while maintaining or improving a peak signal‐to‐noise ratio (PSNR) of 29 dB. The experimental results demonstrate the effectiveness of the proposed model. By applying our modifications, we achieve a notable reduction in model complexity, leading to an improved PSNR of 29.07 dB, up from RFDN's 29.04 dB on a diverse 2 K resolution (DIV2K) dataset. This underscores the potential of our lightweight model to balance computational efficiency and SR quality. Additionally, the proposed work is essential for the metaverse for two key reasons: (1) Enhancing visual quality by adding complex details to textures and objects, making the digital world feel more like reality. (2) Ensuring device compatibility across a range of gadgets, from smartphones to VR headsets, optimizing the metaverse experience for all users. In short, a lightweight single image super‐resolution model for image reconstruction is proposed in this paper.

Recently, in the offline distribution field, as the number of data collection and analysis cases increases by applying IoT devices to kiosks, research on hyper-personalized recommendation systems has become critical. Recommendation systems only work well in some data-rich areas (industries). Therefore, it is unsuitable for kiosk systems with multiple domains and data imbalances, and it is challenging to collect detailed information, such as user reviews and product descriptions. In this article, we propose a context-aware hyper-personalized recommendation system that utilizes context information collected from kiosk IoT devices, minimizes the model size of the kiosk device, and aims for consistent performance and high-recommendation performance in various domains. We also developed effective self-supervised learning to increase data learning efficiency in data imbalance environments. The quality of products recommended by the proposed kiosk recommendation system was evaluated using transactions that occurred in an actual kiosk system. As a result, compared to the existing recommendation system, all performance indicators improved by an average of 20%. When the self-supervised learning method was additionally applied, it improved by an average of 0.8% more. In particular, it shows superior performance regarding the quality of recommended items and resource usage according to users.

Abstract At present, the influence of the surrounding environment on the health of humans is attracting increasing attention, and the indoor environment plays a significant role in people's health. Therefore, it is necessary to effectively monitor indoor environments in real time. With the increase of end‐devices, the rapid growth of data is exerting a pressure on the network bandwidth. To solve this problem, a LoRaWAN monitoring system for indoor environments based on embedded edge computing is proposed here. A hardware architecture of end‐devices and the workflow of the LoRaWAN gateway are designed. In the gateway, edge computing is applied to manage the integrated access device and clean data. The experimental results show that the data reduction rate increased obviously after using edge computing. The environmental data can be transmitted in real time, and the communication efficiency of the system can be improved.

Educators have been calling for reform for a decade. Recent technical breakthroughs have led to various improvements in the semantic web-based education system. After last year's COVID-19 outbreak, development quickened. Many countries and educational systems now concentrate on providing students with online education, which differs greatly from traditional classroom education. Online education allows students to learn at their own pace and the system. As a consequence, we may say that education has become more dynamic. In the educational system, this changing nature makes user demands difficult to identify. Many instructors suggest using machine learning, artificial intelligence, or ontology to improve traditional teaching methods. Due to the lack of survey studies examining and comparing all of the researcher's semantic web-based teaching methodologies, we decided to conduct this survey. This paper's goal is to analyse all available possibilities for semantic web-based education systems that enable new researchers to develop their knowledge.

The development of theindustrial Internet of Thingsparadigm brings forth the possibility of a significant transformation within the manufacturing industry. This paradigm is based on sensing large amounts of data, so that it can be employed by intelligent control systems (i.e.,artificial intelligencealgorithms) eliciting optimal decisions in real time. Ensuring the accuracy and reliability of the intelligent wireless sensing and control system pipeline is crucial toward achieving this goal. Nevertheless, the presence of noise in actual wireless transmission processes considerably affects the quality of the sensed data. Typically, noise and anomalies present in the data are very difficult to distinguish from each other. Conventional anomaly-detection techniques generate many error reports, which cause the control systems to issue incorrect responses that hinder the industrial production. In this article, a novel solution is proposed to denoise data while simultaneously preserving the actual anomalies. The proposed approach operates by measuring both the neighbor and background contrasts in computing a noise score. The trust level of each data point is then calculated through a correlation measure to purge spurious data. Extensive experiments on real datasets demonstrate that the proposed approach yields effective performance, as compared to existing methods, and it meets the requirements of low latency—facilitating the normal operation of the monitored control systems.

Summary At present, human action recognition is a challenging and complex task in the field of computer vision. The combination of CNN and RNN is a common and effective network structure for this task. Especially, we use 3DCNN in CNN part and ConvLSTM in RNN part. We divide the video into multiple temporal segments by average and compress each segment into one feature map by pooling layer. Adding the pooling layer, dropout layer, and batch normalization layer into ConvLSTM is our groundbreaking work. We test our model on KTH, UCF‐11, and HMDB51 datasets and achieve a high accuracy of action recognition.

Internet of Things (IoT) is the utmost assuring framework to facilitate human life with quality and comfort. IoT has contributed significantly to numerous application areas. The stormy expansion of smart devices and their credence for data transfer using wireless mechanics boost their susceptibility to cyberattacks. Consequently, the cybercrime rate is increasing day by day. Hence, the study of IoT security threats and possible corrective measures can benefit researchers in identifying appropriate solutions to deal with various challenges in cybercrime investigation. IoT forensics plays a vital role in cybercrime investigations. This review paper presents an overview of the IoT framework consisting of IoT architecture, protocols, and technologies. Various security issues at each layer and corrective measures are also discussed in detail. This paper also presents the role of IoT forensics in cybercrime investigation in various domains like smart homes, smart cities, automated vehicles, and healthcare. The role of advanced technologies like artificial intelligence, machine learning, cloud computing, edge computing, fog computing, and blockchain technology in cybercrime investigation is also discussed. Lastly, various open research challenges in IoT to assist cybercrime investigation are explained to provide a new direction for further research.

Summary With the growing number of malware, malware analysis technologies need to be advanced continuously. Malware authors use various packing techniques to hide their code from malware detection tools and techniques. The packing techniques are generally used to compress and encrypt executable code in executable files, and the unpacking code is usually embedded in the executable files. Therefore, packed executable files can be executed by itself, and the information associated with packing can be used to analyze and detect malware. Since different packing tools will generate different packed executable files, packing tools can be identified by analyzing packed executable files, and packer identification can reduce malware‐analyzing overheads, and the executable files can even be unpacked. However, most previous studies focused on packing detection using signatures of unpacking code, and these approaches can be avoided by placing unpacking code in other locations or by distributing unpacking code in multiple locations. In this paper, we propose a new packer identification method by analyzing only code sections to extract features of malware generated by different packing tools. Experimental results show that our approach can identify different packing tools with the accuracy of 91.6% on average. Considering packer identification is the harder problem than packing detection, we argue that our approach can contribute to reducing overheads of malware analysis.

Summary Due to the development of mobile transmission techniques, more exciting services can be launched that may change our daily lives in more easy and comfortable ways. The introduction of device‐to‐device (D2D) service in future 5G communication systems can extend the service coverage of intelligent medical treatment from hospitals to anywhere inside the network coverage area, which can save precious time and medical resources. As one of the key techniques in a 5G system, D2D technology has attracted extensive attention from academia and industry due to its system performance improvement, user experience enhancement and service extension. However, the introduction of D2D service will tremendously pollute the system transmission environment due to the induced interference, especially in the downlink direction where the interferer is very strong. This situation brings new challenges for effective and green resource utilization. To solve this problem, a green spectrum resource allocation strategy based on the Hungarian method is proposed in this paper to optimize system spectrum efficiency while considering the fairness among users under the assumption that all resources are fully shared by traditional cellular and newly introduced D2D users. To validate the performance of the proposed algorithm, a system‐level Monte Carlo simulation is also conducted, which shows the favorable performance of the proposed algorithm over the traditional greedy algorithm.

The application of BF techniques in power domain non-orthogonal multiple access (NOMA) schemes allows users to share the same single BF vector for operational reliability. The occurrence of inter-beam interference (IBI) is highly probable in a congested cell (i.e., a cell with high user density and active users). IBI cancellation by using 3D polarized BF in order to enhance the practicability of NOMA systems is the focus of this paper. An IBI cancellation scheme is proposed and evaluations of the spectrum efficiency according to the scenario congestion, as well as of the interference reduction by narrowing the generated beam to a desired beam-width, are presented. The security in the physical (PHY) layer in order to achieve confidential and authentic communication is also an important consideration. The proposed scheme checks the PHY layer security constraints on the number of users served per beam. In simulations, the robustness of the proposed scheme allows the average half power beam-width (HPBW) to be brought to about 20° for different steps in HPBW and for different user densities. Furthermore, depending on the user density, spectrum efficiency gains of approximately 4 bits/s/Hz and 9 bits/s/Hz are achieved by the described IBI cancellation scheme.

The Internet of Things (IoT) that allows connectivity of network devices embedded with sensors undergoes severe data exchange interference as the unlicensed spectrum band becomes overcrowded. By applying cognitive radio (CR) capabilities to IoT, a novel cognitive radio IoT (CR-IoT) network arises as a promising solution to tackle the spectrum scarcity problem in conventional IoT network. CR is a form of wireless communication whereby a radio is dynamically programmed and configured to detect available spectrum channels. This enhances the spectrum utilization efficiency of radio frequency while avoiding interference and overcrowding to other users. Energy efficiency in CR-IoT network must be carefully formulated since the sensor nodes consume significant energy to support CR operations, such as in dynamic spectrum sensing and switching. In this paper, we study channel spectrum sensing to boost energy efficiency in clustered CR-IoT networks. We propose a two-way information exchange dynamic spectrum sensing algorithms to improve energy efficiency for data transmission in licensed channels. In addition, the concern of the energy consumption in dynamic spectrum sensing and switching, we propose an energy efficient optimal transmit power allocation technique to enhance the dynamic spectrum sensing and data throughput. Simulation results validate that the proposed dynamic spectrum sensing technique can significantly reduce the energy consumption in CR-IoT networks.

Security surveillance is critical to social harmony and people's peaceful life. It has a great impact on strengthening social stability and life safeguarding. Detecting anomaly timely, effectively and efficiently in video surveillance remains challenging. This paper proposes a new approach, called S2-VAE, for anomaly detection from video data. The S2-VAE consists of two proposed neural networks: a Stacked Fully Connected Variational AutoEncoder (SF-VAE) and a Skip Convolutional VAE (SC-VAE). The SF-VAE is a shallow generative network to obtain a model like Gaussian mixture to fit the distribution of the actual data. The SC-VAE, as a key component of S2-VAE, is a deep generative network to take advantages of CNN, VAE and skip connections. Both SF-VAE and SC-VAE are efficient and effective generative networks and they can achieve better performance for detecting both local abnormal events and global abnormal events. The proposed S2-VAE is evaluated using four public datasets. The experimental results show that the S2-VAE outperforms the state-of-the-art algorithms. The code is available publicly at https://github.com/tianwangbuaa/.

Recommender systems are increasingly playing an important role in our life, enabling users to find “what they need” within large data collections and supporting a variety of applications, from e-commerce to e-tourism. In this paper, we present a Big Data architecture supporting typical cultural heritage applications. On the top of querying, browsing, and analyzing cultural contents coming from distributed and heterogeneous repositories, we propose a novel user-centered recommendation strategy for cultural items suggestion. Despite centralizing the processing operations within the cloud, the vision of edge intelligence has been exploited by having a mobile app (Smart Search Museum) to perform semantic searches and machine-learning-based inference so as to be capable of suggesting museums, together with other items of interest, to users when they are visiting a city, exploiting jointly recommendation techniques and edge artificial intelligence facilities. Experimental results on accuracy and user satisfaction show the goodness of the proposed application.

Intelligent approaches for security technologiesThese days, processed data size has been increasing and has grown exponentially, that is, not only text data but especially multimedia data with visual and auditory information.Such data may be generated by humans but also by many sources and technical equipment including IoT (Internet of Things) nodes.Intelligent technologies developed for representing the natural way of human thinking have a semantic gap between low level and high level information processing stages.Many research studies for reducing semantic gaps are focused on the representation methods of various

Summary Following the advances in various smart devices, there are increasing numbers of users of social network services (SNS), which allows communication and information sharing in real time without limitations on distance or space. Although personal information leakage can occur through SNS, where an individual's personal details or online activities are leaked, and various financial crimes such as phishing and smishing are also possible, there are currently no countermeasures. Consequently, malicious activities are being conducted through messages toward the users who are in follow or friend relationships on SNS. Therefore, in this paper, we propose a method of assessing follow suggestions from users with less likelihood of committing malicious activities through an information‐driven follow suggestion based on a categorical classification of interests using both the images and text of user posts. We ensure the objectiveness of interest categories by defining these based on DMOZ, which is established by the Open Directory Project. The images and text are learnt using a convolutional neural network, which is one of the machine learning techniques developed with a biological inspiration, and the interests are classified into categories. Users with a large number of posts are defined as certified users, and a database of certified users is established. Users with similar interests are classified, and the similarity distances between certified users and users are measured, and a follow suggestion is generated to the certified user with the most similar interest. Using the method proposed in this paper to classify the interest categories of certified users and users, precisions of 80% and 79.8% were obtained, respectively, and the overall precision was 79.93%, indicating a good classification performance overall. It is expected that the method proposed in this paper can be used to provide follow suggestions of users with less likelihood of malicious activities based on the information posted by the user.

Motivated by the high-interest in increasing the utilization of nongeneral purpose devices in reaching computational objectives with a reduced cost, we propose a new model for scheduling microservices over heterogeneous cloud-edge environments. Our model uses a particular mathematical formulation for describing an architecture that includes heterogeneous machines that can handle different microservices. Since any new model asks for an early risk-analysis of the solution, we improved the CloudSim simulation framework to be suitable for an experiment that includes that kind of systems. In this paper, we discuss two examples of real-life utilizations of our proposed scheduling architecture. For an objective appreciation of the first example, we also include some experimental results based on the developed simulation tool. As a result of our interpretation of the experimental results we find out that some very simple scheduling algorithms may outperform some others in given situations that are frequently present in cloud-edge environments when we are using a microservice-oriented approach.

Including mobile environment, conventional security mechanisms have been adapted to satisfy the needs of users. However, the device environment-IoT-based number of connected devices is quite different to the previous traditional desktop PC- or mobile-based environment. Based on the IoT, different kinds of smart and mobile devices are fully connected automatically via device controller, such as smartphone. Therefore, controller must be secure compared to conventional security mechanism. According to the existing security threats, these are quite different from the previous ones. Thus, the countermeasures applied should be changed. However, the smart device-based authentication techniques that have been proposed to date are not adequate in terms of usability and security. From the viewpoint of usability, the environment is based on mobility, and thus devices are designed and developed to enhance their owners’ efficiency. Thus, in all applications, there is a need to consider usability, even when the application is a security mechanism. Typically, mobility is emphasized over security. However, considering that the major characteristic of a device controller is deeply related to its owner’s private information, a security technique that is robust to all kinds of attacks is mandatory. In this paper, we focus on security. First, in terms of security achievement, we investigate and categorize conventional attacks and emerging issues and then analyze conventional and existing countermeasures, respectively. Finally, as countermeasure concepts, we propose several representative methods.

Several industries in many different domains are looking at deep learning as a way to take advantage of the insights in their data, to improve their competitiveness, to open up novel business possibilities, or to resolve the problem that thought to be impossible to tackle. The large scale of the systems where deep learning is applied and the need of preserving the privacy of the used data have imposed a shift from the traditional centralized deployment to a more collaborative one. However, this has opened up several vulnerabilities caused by compromised nodes and inputs, with traditional crypto primitives and access control models exploited to offer protection means. Providing security can be costly in terms of higher energy consumption, calling for a wise use of these protection means. This paper exploits game theory to model interactions among collaborative deep learning nodes and to decide when using actions to support security enhancements.

There are continuous hacking and social issues regarding APT (Advanced Persistent Threat - APT) attacks and a number of antivirus businesses and researchers are making efforts to analyze such APT attacks in order to prevent or cope with APT attacks, some host PC security technologies such as firewalls and intrusion detection systems are used. Therefore, in this study, malignant behavior patterns were extracted by using an API of PE files. Moreover, the FP-Growth Algorithm to extract behavior information generated in the host PC in order to overcome the limitation of the previous signature-based intrusion detection systems. We will utilize this study as fundamental research about a system that extracts malignant behavior patterns within networks and APIs in the future.

Smart devices have already become a dally necessity of our life since they play a pivot role to record massive amount of Information of our personal life. Current Internet of TWngs era has been seeing more and more people relying on their devices. However, most people are losing patience to set complex passwords to ensure the security for their devices. To avoid forgetting the passwords, people are more willing to choose simple alphanumeric character combinations that are easy for them to remember and convenient to enter. Therefore, their passwords are of high probability to be cracked or exposed. In tWs paper, we study the potential security problems caused by simple and weak passwords, discuss the drawbacks of some conventional works, and propose three schemes to Increase the complexity of simple passwords. Note that our proposals are based on the prediction that the textual passwords are not difficult for users to remember or enter and the proposed schemes can effldently prevent passwords from being cracked or exposed.

Recently, most of context-aware services are trying to exploit the emotional contexts of the target users. The aim of this conceptual paper is to discuss affective lifelogging framework which can recognize the emotions by integrating multimodal information from multiple sources. Moreover, we will mention the open problems on affective lifelogging.

Mobility management has become a core function in internet services and networks as mobile devices have been widely used and their capabilities have dramatically advanced. It is expected that proxy mobile IPv6 (PMIPv6), which is the most prominent solution, will play an important role in supporting these devices’ mobility. To protect PMIPv6 networks, several authentication schemes have been presented. However, due to their static approach, the existing schemes failed to keep a good balance between security and efficiency. Motivated by this, the authors study an adaptive authentication scheme for mobile devices in PMIPv6 networks. In particular, the proposed scheme considers mobile nodes’ context information to decide authentication strength, based on which adaptive authentication is performed. It is shown from the formal security verification and the example study that the proposed scheme is not only correct, but also achieves a good trade‐off between security and efficiency

The most fundamental step in semantic information processing (SIP) is to construct knowledge base (KB) at the human level; that is to the general understanding and conception of human knowledge. WordNet has been built to be the most systematic and as close to the human level and is being applied actively in various works. In one of our previous research, we found that a semantic gap exists between concept pairs of WordNet and those of real world. This paper contains a study on the enrichment method to build a KB. We describe the methods and the results for the automatic enrichment of the semantic relation network. A rule based method using WordNet's glossaries and an inference method using axioms for WordNet relations are applied for the enrichment and an enriched WordNet (E-WordNet) is built as the result. Our experimental results substantiate the usefulness of E-WordNet. An evaluation by comparison with the human level is attempted. Moreover, WSD-SemNet, a new word sense disambiguation (WSD) method in which E-WordNet is applied, is proposed and evaluated by comparing it with the state-of-the-art algorithm.