نشریه مهندسی برق و مهندسی کامپیوتر ایران
سال شانزدهم شماره 3 (پیاپی 48، پاییز 1397)

Due to the high-speed of information generation and the need for data-knowledge conversion, there is an increasing need for data mining algorithms. Clustering is one of the data mining techniques, and its development leads to further understanding of the surrounding environments. In this paper, a dynamic and scalable solution for clustering mixed big data with a lack of data is presented. In this solution, the integration of common distance metrics with the concept of the closest neighborhood, as well as a kind of geometric coding are used. There is also a way to recover missing data in the dataset. By utilizing parallelization and distribution techniques, multiple nodes can be scalable and accelerated. The evaluation of this solution is based on speed, precision, and memory usage criteria compared to other ones.

Memetic algorithm (MA) is a kind of evolutionary algorithms (EAs) that searches the problem solving space using local search and global search. The balance between global search and local search is one of the key issues in this algorithm. In this paper a new model is proposed, called GALA2. This model is combined of genetic algorithm (GA) and object migration automata (OMA), which is a kind of fixed-structure learning automaton. In the proposed model, global search is performed by genetic algorithm and local learning is performed by learning automata. In this model, the Lamarckian and Baldwinian learning models have been used to increase the convergence rate and avoidance of premature convergence, simultaneously. In this evolutionary model, chromosomes are represented by object migration automata for the purpose of using positive effects of evolution and local learning. In order to show the superiority of the proposed model, GALA2 is used to solve the graph isomorphism problem.

Cognitive radio as a key technology is taken into consideration widely to cope with the shortage of spectrum in wireless networks. One of the major challenges to realization of CR networks is security. The most important of these threats is primary user emulation attack, thus malicious user attempts to send a signal same as primary user's signal to deceive secondary users and prevent them from sending signals in the spectrum holes. Meanwhile, causing traffic in CR network, malicious user obtains a frequency band to send their information. In this thesis, a method to identify primary user emulation attack is proposed. According to this method, primary users and malicious users are distinguished by clustering. In this method, the number of active users is recognized in the CR network by clustering. Indeed, by using Dirichlet process mixture model classification based on the Bayesian Nonparametric method, primary users are clustered. In addition, to achieve higher convergence rate, Chinese restaurant process method to initialize and non-uniform sampling is applied to select clusters parameter.

Wireless sensor networks (WSNs) consist of thousands of small nodes. The small and inexpensive parts of these nodes have led to their widespread use in various fields. However, these networks have constraints on energy consumption, processing resources, and storage which have caused many studies to find solutions to reduce these constraints. In recent years, with the advent of the concept of Fog computing, many new and effective solutions are represented for routing in wireless sensor networks. Since in WSNs it is important to save alive nodes and reduce the energy consumption of nodes, fog computing is useful for this purpose. In most WSN routing protocols, the best way to send data to cluster heads and the base station is the major part of their studies. In the new protocols, the Fog computing have been used to find the best way. In these methods, we have seen decreasing energy consumption and increasing network lifetime. In this paper, we represent a fog-based algorithm for routing in WSNs. According to the simulation results, the proposed protocol improved energy consumption by 9% meanwhile the number of alive nodes is increased by 74%, compared to the reviewed method.

The image observer usually wants to receive the message and the main subject of the image in the shortest time. Hence, assuming there is useful information in the salient regions, the human vision system unconsciously guides visual attention towards them. This assumption is not always correct in practice, and in some cases, salient regions merely cause visual distractions. Therefore, in different applications, a mechanism is needed to identify these regions. To prevent from distracting observer’s attention from the main subject, these regions are eliminated. Furthermore, neglecting these regions could be of considerable assistance to the methods that function base on salient regions recognition. So, in this paper, Based on the methods of the class imbalance challenge each segment of training images in the dataset is a partition to 9 classes according to the relevant mask in the dataset, that the number of each class is proportional to its disturbance intensity. Then, segment-based features are extracted and determining the class of each segment is determined according to WTPTSSR method, which is based on the Sparse Coding and Representation system.Finally, in order to precisely analyzing the proposed method and comparing it to other approaches, four analysis criteria with different performances are presented. According to results, despite being time-consuming, the proposed method has a higher accuracy than the previous ones.

Nowadays the network science has been attracted many researchers from a wide variety of different fields and many problems in engineering domains are modelled through social networks measures. One of the most important problems in social networks is the prediction of evolution and structural behavior of the networks that is known as link prediction problem in the related literature. Nowadays people use multiple and different social networks simultaneously and it causes to demonstrate a new domain of research known as heterogenous social networks. There exist a few works on link prediction problem on heterogenous networks. In this paper, first a novel similarity measure for users in heterogenous networks is defined. Then a novel link prediction algorithm is described through a supervised learning approach which is consisted by the generated features from the introduced similarity measures. We employ the standard evaluation criteria for verification of the proposed approach. The comparison of the proposed algorithm to the other well-known earlier works showed that our proposed method has better performance than the other methods based on testing on several network datasets.

Coverage of an area, with one or multiple sensors, is one of the fundamental challenges in wireless sensor networks. Since a sensor life span is limited and reliable data is of great importance, sensitive applications like fire\leakage alarm systems, intrusion detection, etc. need multiple sensors to cover the region of interest, which is called K-coverage. Most of the studies that have been carried out on K-coverage evaluation have used binary sensing model. In this paper, we propose a grid-based K-coverage evaluation technique using probabilistic sensing model to increase evaluation accuracy and decrease evaluation time. The proposed technique is implemented using NS-2 simulator, and its results are compared to probabilistic perimeter-based and binary grid-based techniques. The results indicate that the proposed technique improved accuracy by 14% and 24% compared to the mentioned techniques respectively. It also shows 7% decrease in evaluation time compared to probabilistic perimeter-based technique.

Semi-supervised metric learning has attracted increasing interest in recent years. They enforce smoothness label assumption on the manifold. However, they suffer from two challenges: (1) since data in each class lies on one manifold and the similarity between classes leads the intersection between manifolds, the smoothness assumption on the manifold is violated in intersecting regions. (2) 1NN classifier, which is applied for predicting the label of classes in metric learning methods, is suffered from the rare of labeled data and has not suitable accuracy. In this paper, a novel method for learning semi-supervised metric in the stratified space has been proposed that exploit the prior knowledge, which is the smoothness assumption on each manifold, more accurate than existing methods. In the metric learning stage, it doesn’t apply smoothness assumption on the intersecting regions and in the classification stage, labeled data in the interior regions of manifolds are extended based on the smoothness assumption. The different behavior of the Laplacian of piecewise smooth function on stratified space is exploited for the distinction of the intersecting regions from interior regions of manifolds. The results of experiments verify the improvement of the classification accuracy of the proposed method in the comparison with other methods.