Vol. 42 · No. 2

In heterogeneous information network (HIN)-based applications, the existing methods usually use Heterogeneous Graph Neural Networks (HGNN) to handle some complex tasks. However, these methods still have some shortcomings: 1) they manually pre-select some meta-paths and thus some important ones are missing, while the missing ones still contains the information and features of the node in the entire graph structure; and 2) they have no high interpretability since they do not consider the logical sequences in an HIN. In order to deal with them, we propose ASIAM-HGNN: a heterogeneous graph neural network combined with the automatic selection and interpretable aggregation of meta-path instances. Our model can automatically filter important meta paths for each node, while preserving the logical sequence between nodes, so as to solve the problems existing in other models. A group of experiments are conducted on real-world datasets, and the results demonstrate that the models learned by our method have a better performance in most of task scenarios.

In this paper, a novel approach has been proposed for hiding sensitive association rules based on the affinity between the frequent items of the transaction. The affinity between the items is defined as Jaccard similarity. This work proposes five algorithms to ensure the minimum side-effects resulting after applying sanitization algorithms to hide sensitive knowledge. Transaction affinity has been introduced which is calculated by adding the affinity of frequent items present in the transaction with the victim-item (item to be modified). Transactions are selected either by increasing or decreasing value of affinity for data distortion to hide association rules. The first two algorithms, MaxaffinityDSR and MinaffinityDSR, hide the sensitive information by selecting the victim item as the right-hand side of the sensitive association rule. The next two algorithms, MaxaffinityDSL and MinaffinityDSL, select the victim item from the left-hand side of the rule whereas the Hybrid approach picks the victim item from either the left-hand side or right-hand side. The performance of proposed algorithms has been evaluated by comparison with state-of-art methods (Algo 1.a and Algo 1.b), MinFIA, MaxFIA and Naive algorithms. The experiments were performed using the dataset generated from IBM synthetic data generator, and implementation has been performed in R language.

With the rapid increase of Internet of Things (IoT) devices and applications, the ordinary cloud computing paradigm soon becomes outdated. Fog computing paradigm extends services provided by a cloud to the edge of network in order to satisfy requirements of IoT applications such as low latency, locality awareness, low network traffic, mobility support, and so forth. Task scheduling in a Cloud-Fog environment plays a great role to assure diverse computational demands are met. However, the quest for an optimal solution for task scheduling in the such environment is exceedingly hard due to diversity of IoT applications, heterogeneity of computational resources, and multiple criteria. This study approaches the task scheduling problem with aims at improving service quality and load balancing in a merged computing system of Edge-Fog-Cloud. We propose a Multi-Objective Scheduling Algorithm (MOSA) that takes into account the job characteristics and utilization of different computational resources. The proposed solution is evaluated in comparison to other existing policies named LB, WRR, and MPSO. Numerical results show that the proposed algorithm improves the average response time while maintaining load balancing in comparison to three existing policies. Obtained results with the use of real workloads validate the outcomes.

Neural networks are increasingly used in recognition problems, including static and moving images, sounds, etc. Unfortunately, the selection of optimal neural network architecture for a specific recognition problem is a difficult task, which often has an experimental nature. In this paper we present the use of evolutionary algorithms to obtain optimal architectures of neural networks used for audio sample classification. We extend the Pytorch DNN Evolution tool implementing co-evolutionary algorithms which create groups of neural networks that solve a given problem with a certain accuracy, with the support for problems in which training data consists of audio samples. In this paper we use the co-evolutionary approach to solve a sample sound classification problem. We describe how the sound data was prepared for processing with the use of the Mel Frequency Cepstral Coefficients (MFCC). Next we present the results of experiments conducted with the AudioMnist dataset. The obtained neural network architectures, whose classification accuracy is comparable to the classification accuracy attained by the AlexNet neural network, and their implications are discussed.

CloudIoT is a new paradigm, which has emerged as a result of the combination of Cloud Computing (CC) and the Internet of Things (IoT). It has experienced a growing and rapid development, and it has become more popular in information and technology (IT) environments because of the advantages it offers. However, due to a strong use of this paradigm, especially in smart cities, the problem of imbalance load has emerged. Indeed, to satisfy the needs of the user, the intelligent objects send the collected data to the virtual machines (VMs) of the cloud in order to be processed. So, it is necessary to have an idea about the load of its VM. Thus, the problem of load balancing between VMs is strongly related to the technique used for the VMs selection. To tackle this problem, we propose in this paper a task scheduler called Scheduler Genetic Grasshopper Algorithm (SGGA). It allows to ensure a dynamic load balancing, as well as the optimization of the makespan and the resource usage. Our proposed SGGA is based on the combination of Genetic Algorithm (GA) and Grasshopper Optimization Algorithm (GOA). First, the tasks sent by the IoTs are mapped to the VMs in order to build the initial population, then SGGA performs the genetic algorithm, which has expressed a considerable performance. However, the weakness of the GA is marked by its heaviness caused by the mutation operator, especially when the number of tasks increases. Because of this insufficiency, we have replaced the mutation operator with the grasshopper optimization algorithm. The results of the experiments show that our approach (SGGA) is the most efficient, compared to the recent approaches, in terms of the response time to obtain the optimal solution, makespan, throughput, an average resource utilization rate and the hypervolume indicator.

The Transformer has been widely used for many tasks in NLP before, but there is still much room to explore the application of the Transformer to the image domain. In this paper, we propose a simple and efficient hybrid Transformer framework, CTransNet, which combines self-attention and CNN to improve medical image segmentation performance. Capturing long-range dependencies at different scales. To this end, this paper proposes an effective self-attention mechanism incorporating relative position information encoding, which can reduce the time complexity of self-attention from O(n2) to O(n), and a new self-attention decoder that can recover fine-grained features in encoder from skip connection. This paper aims to address the current dilemma of Transformer applications: i.e., the need to learn induction bias from large amounts of training data. The hybrid layer in CTransNet allows the Transformer to be initialized as a CNN without pre-training. We have evaluated the performance of CTransNet on several medical segmentation datasets. CTransNet shows superior segmentation performance, robustness, and great promise for generalization to other medical image segmentation tasks.

With the progress of society and the improvement of living standards, people pay more and more attention to personal health, and WITMED (Wise Information Technology of med) has occupied an important position. The relationship prediction work in the medical field has high requirements on the interpretability of the method, but the relationship between medical entities is complex, and the existing methods are difficult to meet the requirements. This paper proposes a novel medical information relation extraction method MGCN, which combines contextual information to provide global interpretability for relation prediction of medical entities. The method uses Co-occurrence Graph and Graph Convolutional Network to build up a network of relations between entities, uses the Open-world Assumption to construct potential relations between associated entities, and goes through the Knowledge-aware Attention mechanism to give relation prediction for the entity pair of interest. Experiments were conducted on a public medical dataset CTF, MGCN achieved the score of 0.831, demonstrating its effectiveness in medical relation extraction.

Intuitionistic fuzzy graphs are extensions of fuzzy graphs that preserve the dualism characteristics of fuzzy graphs and have a stronger capacity to describe ambiguity in actual decision-making issues than fuzzy graphs. In this research paper, the Laplacian energy and correlation coefficient of intuitionistic fuzzy graphs are computed for finding group decision-making problems that are supported by intuitionistic fuzzy preference relations. We propose a novel method for calculating establishments' comparative position loads by manipulating the undecided corroboration of IFPR and the correlation coefficient of one personality IFPR to the other items. As a result, we comprehend a large number of establishments in the detailed IFPR and devise a correlation coefficient process to investigate the significance of alternatives and the best of the alternatives. Finally, we present a collaborative decision-making technique in a money-investing scheme, and that idea may be devised in disparate beneficial investing schemes.

It is very difficult to accomplish the 3D reconstruction of the clothed human body from a single RGB image, because the 2D image lacks the representation information of the 3D human body, especially for the clothed human body. In order to solve this problem, we introduced a priority scheme of different body parts spatial information and proposed PointHuman network. PointHuman combines the spatial feature of the parametric model of the human body with the implicit functions without expressive restrictions. In PointHuman reconstruction framework, we use Point Transformer to extract the semantic spatial feature of the parametric model of the human body to regularize the implicit function of the neural network, which extends the generalization ability of the neural network to complex human poses and various styles of clothing. Moreover, considering the ambiguity of depth information, we estimate the depth of the parameterized model after point cloudization, and obtain an offset depth value. The offset depth value improves the consistency between the parameterized model and the neural implicit function, and accuracy of human reconstruction models. Finally, we optimize the restoration of the parametric model from a single image, and propose a depth perception method. This method further improves the estimation accuracy of the parametric model and finally improves the effectiveness of human reconstruction. Our method achieves competitive performance on the THuman dataset.

In cyberspace, there exists a prevalent problem that heavily occurs to web application databases and that is the exploitation of websites by using SQL injection attacks. This kind of attack becomes more difficult when it comes to blind SQL vulnerabilities. In this paper, we will first make use of this vulnerability, and subsequently, we will build an ontology (OBSQL) to address the detection of the blind SQL weakness. Therefore, to achieve the exploitation, we reproduce the attacks against a website in production mode. We primarily detect the presence of the vulnerability, after we use our tools to abuse it. Last but not least, we prove the importance of applying ontology in cybersecurity for this matter. The mitigation techniques in our ontology will be addressed in our future work.

With the rapid spread of e-commerce and e-payment, the increasing number of people choose online shopping instead of traditional buying way. However, the malicious user behaviors have a significant influence on the security of users' accounts and property. In order to guarantee the security of shopping environment, a method based on Complex Event Process (CEP) and Colored Petri nets (CPN) is proposed in this paper. CEP is a data-driven technology that can correlate and process a large amount of data according to Event Patterns, and CPN is a formal model that can simulate and verify the specifications of the online shopping processes. In this work, we first define the modeling scheme to depict the user behaviors and Event Patterns of online shopping processes based on CPN. The Event Patterns can be constructed and verified by formal methods, which guarantees the correctness of Event Patterns. After that, the Event Patterns are translated into Event Pattern Language (EPL) according to the corresponding algorithms. Finally, the EPLs can be inserted into the complex event processing engine to analyze the users' behavior flows in real-time. In this paper, we validate the effectiveness of the proposed method through case studies.