مجله رایانش نرم و فناوری اطلاعات
سال نهم شماره 2 (تابستان 1399)

Due to the development of the internet and the diversity of information, decision making in various fields has been faced different challenges. Recommender systems by identifying users interests, data filtering, and data management offer personalized services to users. This is beneficial for marketing and user satisfaction. Collaborative Filtering (CF) is one of the most successful methods of recommender system. CF is based on the similarity between users. We argue that similarity is a fuzzy notion and we get more realistic results in recommender systems by using fuzzy logic. Fuzzy logic is an effective way to identify ambiguities and uncertainty in measuring the similarity of items and users. In this paper, we present a new Fuzzy Similarity Measure, called FSM, for CF recommender systems which is based on popularity and significance. To evaluate the contribution of this work, we use MAE, F1, recall, and precision. Using the proposed fuzzy similarity measure, FSM, we obtain a F1 value equal to 0.6550, which outperforms the PIP and NHSM respectively by %17 and %20. Also, the MAE value based on the proposed fuzzy similarity measure is equal to 0.4604, which outperforms the NHSM by %5. We have also observed an improvement in recall and precision using the proposed similarity measure.

Hybrid systems consist of both continuous and discrete parts. These systems include several different components. A fault in one of these components can be activated and propagate to other components. Due to the advancement of technology and intelligent systems such as driverless car, health control devices and automated factories, the occurrence of a fault in one component and its propagation to other components can lead to financial and human-life losses. It is necessary to design a fault propagation model before construction of a system. With such a model, we can observe the propagation of the effects of a fault in a component to other components, before the construction of the system. It is also possible to identify critical components of the system. In this paper, a method for modeling fault propagation based on stochastic activity networks is presented. Based on this model, it is possible to identify the critical points of the system, the effect of different components on each other and the component failure behavior. The proposed model has been applied in a case study (an aircraft fuel system), and its simulation and quantitative results are presented in this paper.

Nowadays, the rapid growth of the use of information technology and computer networks has increased the transfer of information in a digital form. For this reason, the protection of data has become one of the most important challenges in this field. Watermarking is introduced as one of the newest and most important techniques for data protection. Audio watermarking is considered to be the most challenging type of watermarking due to the nature of audio files. The most extraction methods used in audio watermarking algorithms, which mainly use non-intelligent techniques based on the reverse of embedding rules for extraction phase of audio watermarking, often they are not able to extract watermarking exactly and have a lot of errors in extracting. Our proposed solution to solve this problem is to use an intelligent algorithm to extract the watermark. The purpose of this article is to provide a method that covered the weakness of non-intelligent extraction methods using trained machine learning classifier and helped to improve system performance. For the embedding operation, the Lifting Wavelet Transform (LWT) has been used in the proposed method; in the extraction operation, the Support Vector Machine (SVM) classifier is also used. The trained classifier is able to detect the effects of various attacks on the audio files and consequently, intelligent and precise extraction of watermark. The results of various experiments under different conditions indicate that this intelligent method has achieved appropriate imperceptibility and high capacity along with high robustness.

Transfer learning and domain adaptation are effective solutions for performance improvement of image classifiers where the source domain (training set) and target domain (test set) have substantial probability distribution differences. In fact, collecting input data in various conditions (such as lighting or temperature), different equipment with variable characteristics (such as number of input ports or resolution quality) and different views (such as dimensions or environment) results domain shift problem. Semi-supervised domain adaptation is a leading solution for domain shift problem, where the source domain and a small part of target domain are labeled. In this paper, we propose KErnelized Domain Adaptation and balanced distributions alignment (KEDA) to adapt the source and target domains in a semi-supervised manner. KEDA preserves the topology of domains via creating a Laplacian matrix and similarity and dissimilarity views. Moreover, KEDA adapts the regularized conditional and marginal distributions across domains. Ultimately, the sum of these solutions leads to a good classification function for labeling the unlabeled images. The proposed method is compared with the state-of-the-art methods of domain adaptation on Office-Caltech-10, Digits, Pie, and Coil datasets where results show considerable performance improvement of our proposed method.

Abstract- In this paper, a method is presented for search space reduction in Farsi Printed Sub words recognition. First 10 simple features from sub word are extracted. By using the concept of quantization, These features are quantized according to the interval changes of each feature in training data, and are converted to integers. A score is given to every class, using each feature and its distance to corresponding feature of each training sample. By applying all features, each class has a score per feature. A final score is obtained, by fusion of these scores using algebra operations, for each class. Search space is reduced using sorting of final scores and selection of some sub words with more scores. For fusion of scores, sum, prod, max, min and weighted sum operations are used. The weighted sum method, which Optimized weights are obtained by particle swarm optimization (PSO), has given the best response.

In this paper, an efficient approach for dynamic congestion management and traffic control of reconfigurable network-on-chip (NoC) is proposed. In this approach that called DyCM, the objectives of improving the parameters of delay, area, throughput and power consumption are pursued. Using he proposed scheme in the router control unit, several virtual channels on each port based on the network traffic are supported. Dynamic and non-uniform allocation of virtual channels, as well as management of turning on and off virtual channels, ports and routers, considerably reduce power consumption of the routers, and also result in increasing throughput. The overhead of the proposed design is increasing the router area, due to the complexity of the power management unit. To reduce the area overhead and to increase the speed of the router, we have reduced the number of pipeline stages in the proposed router. The evaluations show that the DyCM offers higher performance in terms of speed, power consumption and network throughput than recent related architectures.

air pollution modeling is essential for environmental management. we attempted to select the optimum interpolation method for modeling air pollution in Tehran using some selected methods and proposed method. For this purpose, different interpolation methods such as inverse distances interpolation, polynomials, K nearest neighbors and various existing kriging methods were employed. Finally, by evaluating all the methods, the optimum interpolation method for air pollution modeling in Tehran was proposed and implemented. Among the above-mentioned methods, Kriging has had the best results. Then, using genetic optimization and particle swarm optimization methods, the clustering intervals of the Kriging method were optimized in both regular and irregular intervals to determine the optimal number of clusters and the distance between the clusters at regular intervals in the conventional methods. A mathematical variogram was used, in the irregular interval method. the same experimental variogram employed and the aim was to minimize the error of each of these methods. Finally, the error of the regular intervals with the optimal cluster numbers and spacing was less than those of all the mentioned methods. To test the proposed model, this model has been implemented for all months of the year (March 2016 – March 2017), with an average %64 accuracy improvement compared to that of the regular kriging method. The least error in the methods for interpolation of air pollution has been at 4 (μg / m3), while an error of 1.8 (μg / m3) in this study was achieved which resulted to the %55 accuracy improvement of the interpolation of air pollution, approximately. As a result, in this research, the best interpolation method has been presented in relation to previous studies for air pollution modeling. the proposed method can be applied as a complementary method for air pollution modeling where information scarcity exists in air pollution contamination.

Various systems such as an education system should adapt with the increasing development of technology and use it for pursuing their purposes. Hence, the concept of EE or web based education is considered in the recent years. Beside the various advantages of this highly useful model of education, limitations such as absence of proper infrastructure, improper resource efficiency, and hardware and software resource management cost, exist. This paper aims to tackle the mentioned challenges through benefitting from the advantages of CC. For this purpose, after introducing CC technology, its advantages in EE System (EES) and the related works are given. The proposed intelligent architecture for a cloud-based EES is presented afterwards which is applicable for various courses in different educational systems. A comprehensive application layer and an intelligent and highly flexible evaluation mechanism are some of the advantages of this architecture that result in higher user satisfaction and performance improvement. Other benefits include accurate insight to the system user roles, their accessible sections, and accordingly considering the required sections in the cloud management layer. After examining different methods of implementing the proposed architecture, it has been evaluated against customary EES and also the recently presented cloud-based architectures. Finally, aside from showing its superiority, the proposed steps towards transition to the proposed superior architecture and also its limitations are discussed.

Breast cancer is the most common cancer among women and the existence of a precise and reliable system for the diagnosis of benign or malignant of this cancer is essential. Nowadays, using the results of needle aspiration cytology, data mining and machine learning techniques, early diagnosis of breast cancer can be done with greater accuracy. In this study, we propose a method consisting of two steps: in the first step, to eliminate the less important features, logistic regression has been used to select more important features. In the second step, the Support Vector Machine (SVM) classification algorithm has been used with three different kernel functions for the diagnosis of benign and malignant samples. To evaluate the performance of the proposed method, two data sets, WBCD and WDBC have been used with investigation of several metrics such as precision, the Area Under the ROC (AUC), true positive rate, false positive rate, accuracy and the F-measure. The results show that using the logistic regression method, it is possible to select the more efficient features, such that the proposed method reaches 98.69% in terms of classification accuracy.

Abstract- Edge detection is carried out using high-quality and real-time structural forests method. Nevertheless, in the output of this method, edges with a width of more than one pixel as well as unreal edges that do not belong to any boundary can be seen. Applying thresholds to this output cannot also eliminate all of these disadvantages and sometimes leads to the removal of edge pixels resulting in a loss of performance. In this paper, by presenting the extended structural forests method based on contour grouping and reasoning, while eliminating the disadvantages of the basic structural forests edge detector and improving its performance, a precise and high-quality boundary detector is achieved. By using this method and according to the F-measure criterion, the performance of edge detection is improved 2.26% in total. The proposed algorithm can be used as an effective preprocessing stage in many image processing functions by providing accurate boundaries with a width of one pixel.

In this paper, we present a generalized mathematical model of propagation of malware objects in computer networks which is based on the epidemic modeling of the virus in biological phenomena. In this model, we partition the network computers into five groups; Vulnerable population, Exposed population, Infectious population, Quarantined population, Recovered population. Furthermore, we can model the changes in the status of computers at a malware attack by an ordinary differential equations theory based on the general performance of anti- malware software. A nonlinear differential equation system is proposed for this model, then stability and instability of the network (in a malware attack) are investigated by qualitative theories. In addition, we show that in which conditions the network has a normal function or a potential disruption when a malware attack occurs. Also, our theoretical results are validated by the aid of numerical simulations which are based on real data of an anti-malware software in a multi-worm attack.

Correct detection and classification of power quality disturbances is an essential issue in power systems. In this paper, an intelligent method is proposed to detect the power quality disturbances. This method is based on an identity vector framework that produces a fixed-length vector for each perturbation. In the first step of the proposed pipeline, the discrete wavelet transform is used to analyze power quality events and extract the features of each input signal, and then the identity vector is built using the approximation coefficients. After applying some normalization on the obtained identity, it is classified using a support vector machine classifier. In order to evaluate the proposed method, twelve types of disturbances have been synthesized and the efficiency of the proposed system is investigated using them. In addition, to verify the robustness of the proposed approach towards the noise, the synthesized signals are contaminated with white gaussian noise with different SNR values, 30 dB, 40 dB and 50 dB. The results of the experiments demonstrate the efficiency of the proposed method for the classification of power quality signals with an accuracy of 99.2%.

Hybrid cache architectures are promising solutions for designing efficient I/O caching schemes. Oneof the major challenges of hybrid caches is the management of read miss data pages. As such pages are non-critical in term of data reliability, they can be buffered in volatile memories such as DRAM as an Endurance Friendly (EF) solution. Low capacity of DRAM, however, provides small hit ratio for these pages. A state-of-the-art solution has proposed a State-Machine Based Insertion (SMBI) scheme in order to improve cache performance by inserting read miss data pages in a Solid-State Drive (SSD), based on a static time periods. Nevertheless, the static approach of SMBI leads to considerable cache lifetime degradation compared to EF policy. In this paper, we propose Conservative State-Machine based Insertion (CSMI) scheme to improve I/O cache lifetime. CSMI dynamically adjusts the time that state-machine exploits EF approach based on workload behavior. To this end, CSMI employs a Conservative Factor (CF), which multiplies the number of periods that EF policy is exploited in case of low hit ratio. Our evaluations show that CSMI reduces the number of SSD writes up to 33.97%, compared to state-of-the art solution with negligible performance overhead (average of 0.06%).

Transmitting voice over wireless LAN is a useful cost-effective technique that provides vocal conversation service in broadband wireless networks according to the 802.11 standards. In this service, SIP is responsible to provide sessions. Recent researches show that a call admission control algorithm that controls SIP sessions can guarantee the quality of service parameters in transmitting vocal data. State- of-the-art algorithms, deciding at wireless access-points and altering MAC sub-layer parameters, have high complexity and are not efficient in providing quality of service. Therefore, in this paper, we present a new algorithm that decides on accepting or rejecting new calls in a considerably small time. The Proposed algorithm considers dynamic conditions of the network including channel busy rate and the ratio of real-time traffic of channel and makes no change in MAC sub-layer parameters. Testing the proposed method on a real substrate proves its superiority as compared to the best of recent methods in providing requested level of quality of service in terms of call response time, the average duration of calls, packet loss ratio and the delay of real-time traffic.