نشریه مهندسی برق و مهندسی کامپیوتر ایران
سال نوزدهم شماره 4 (پیاپی 67، زمستان 1400)

Deployment applications of internet of things (IoT) in fog infrastructure as cloud complementary leads effectively computing resource saving in cloud infrastructure. Recent research efforts are investigating on how to better exploit fog capabilities for execution and supporting IoT applications. Also, the distribution of an application’s components on the possible minimum number of fog nodes for the sake of reduction in power consumption leads degradation of the service reliability level. In this paper, a hybrid meta-heuristic algorithm based on cuckoo search algorithm is presented for static deployment the components of IoT applications on fog infrastructure in the aim of trade-off between efficient power usage, reduction in the effect of one point of failure and boosting the application reliability against failure. The results of simulations show that the proposed approach in this paper reduces the power consumption of fog network and meets the quality of service requirement of IoT application with the high reliability level.

This paper investigates the problem of data gathering in rechargeable Wireless Sensor Networks (WSNs). The low energy harvesting rate of rechargeable nodes necessitates effective energy management in these networks. The existing schemes did not comprehensively examine the important aspects of energy-aware data gathering including sleep scheduling, and energy-aware clustering and routing. Additionally, most of them proposed greedy algorithms with poor performance. As a result, nodes run out of energy intermittently and temporary disconnections occur throughout the network. In this paper, we propose an energy-efficient data gathering algorithm namely Energy-aware Data Gathering in Rechargeable wireless sensor networks (EDGR). The proposed algorithm divides the original problem into three phases namely sleep scheduling, clustering, and routing, and solves them successively using particle swarm optimization algorithm. As derived from the simulation results, the EDGR algorithm improves the average and standard deviation of the energy stored in the nodes by 17% and 5.6 times, respectively, compared to the previous methods. Also, the packet loss ratio and energy consumption for delivering data to the sink of this scheme is very small and almost zero

Nowadays, clustering methods have received much attention because the volume and variety of data are increasing considerably.The main problem of classical clustering methods is that they easily fall into local optima. Meta-heuristic algorithms have shown good results in data clustering. They can search the problem space to find appropriate cluster centers. One of these algorithms is gray optimization wolf (GWO) algorithm. The GWO algorithm shows a good exploitation and obtains good solutions in some problems, but its disadvantage is poor exploration. As a result, the algorithm converges to local optima in some problems. In this study, an improved version of gray optimization wolf (GWO) algorithm called 4-gray wolf optimization (4GWO) algorithm is proposed for data clustering. In 4GWO, the exploration capability of GWO is improved, using the best position of the fourth group of wolves called scout omega wolves. The movement of each wolf is calculated based on its score. The better score is closer to the best solution and vice versa. The performance of 4GWO algorithm for the data clustering (4GWO-C) is compared with GWO, particle swarm optimization (PSO), artificial bee colony (ABC), symbiotic organisms search (SOS) and salp swarm algorithm (SSA) on fourteen datasets. Also, the efficiency of 4GWO-C is compared with several various GWO algorithms on these datasets. The results show a significant improvement of the proposed algorithm compared with other algorithms. Also, EGWO as an Improved GWO has the second rank among the different versions of GWO algorithms. The average of F-measure obtained by 4GWO-C is 82.172%; while, PSO-C as the second best algorithm provides 78.284% on all datasets.

In recent years, the issue of vulnerability of machine learning-based models has been raised, which shows that learning models do not have high robustness in the face of vulnerabilities. One of the most well-known defects, or in other words attacks, is the injection of adversarial examples into the model, in which case, neural networks, especially deep neural networks, are the most vulnerable. Adversarial examples are generated by adding a little purposeful noise to the original examples so that from the human user's point of view there is no noticeable change in the data, but machine learning models make mistakes in categorizing the data. One of the most successful methods for modeling data uncertainty is Gaussian processes, which have not received much attention in the field of adversarial examples. One reason for this could be the high computational volume of these methods, which limits their used in the real issues. In this paper, a scalable Gaussian process model based on random features has been used. This model, in addition to having the capabilities of Gaussian processes for proper modeling of data uncertainty, is also a desirable model in terms of computational cost. A voting-based process is then presented to deal with adversarial examples. Also, a method called automatic relevant determination is proposed to weight the important points of the images and apply them to the kernel function of the Gaussian process. In the results section, it is shown that the proposed model has a very good performance against fast gradient sign attack compared to competing methods.

Software-defined networking is a new network architecture which separates the control layer from the data layer. In this approach, the responsibility of the control layer is delegated to the controller software to dynamically determine the behavior of the entire network. It results in a flexible network with centralized management in which network parameters can be well controlled. Due to the increasing number of users, the emergence of new technologies, the explosive growth of network traffic, meeting the requirements of quality of service and preventing underload or overload of resources, load balancing in software-based networks is of substantial importance. Load imbalance increases costs, reduces scalability, flexibility, efficiency, and delay in network service. So far, a number of solutions have been proposed to improve the performance and load balancing in the network, which take into account different criteria such as power consumption and server response time, but most of them do not prevent the system from entering the load imbalance mode and the risks of load imbalance. In this paper, a predictive load balancing method is proposed to prevent the system from entering the load imbalance mode using the Extreme Learning Machine (ELM) algorithm. The evaluation results of the proposed method show that in terms of controller processing delay, load balance and response time, it performs better than CDAA and PSOAP methods.

One of the challenges of high dimensional outlier detection problem is the curse of dimensionality which irrelevant dimensions (features) lead to hidden outliers. To solve this problem, some dimensions that contain valuable information to detect outliers are searched to make outliers more prominent and detectable by mapping the dataset into the subspace which is constituted of these relevant dimensions/features. This paper proposes an outlier detection method in high dimensional data by introducing a new locally relevant subspace selection and developing a local density-based outlier scoring. First, we present a locally relevant subspace selection method based on local entropy to select a relevant subspace for each data point due to its neighbors. Then, each data point is scored in its relevant subspace using a density-based local outlier scoring method. Our adaptive-bandwidth kernel density estimation method eliminates the slight difference between the density of a normal data point and its neighbors. Thus, normal data are not wrongly detected as outliers. At the same time, our method underestimates the actual density of outlier data points to make them more prominent. The experimental results on several real datasets show that our local entropy-based subspace selection algorithm and the proposed outlier scoring can achieve a high accuracy detection rate for the outlier data.

Control of areas and locations and motion sensors in the Internet of Things requires continuous control to detect human activities in different situations, which is an important challenge, including manpower and human error. Permanent human control of IoT motion sensors also seems impossible. The IoT is more than just a simple connection between devices and systems. IoT information sensors and systems help companies get a better view of system performance. This study presents a method based on deep learning and a 30-layer DNN neural network for detecting human activity on the Fordham University Activity Diagnostic Data Set. The data set contains more than 1 million lines in six classes to detect IoT activity. The proposed model had almost 90% and an error rate of 0.22 in the evaluation criteria, which indicates the good performance of deep learning in activity recognition.

In recent years, boxwood dieback has become one of the essential concerns of practitioners and managers of the natural resources of the country. To control the expansion of the factors contributing to the dieback of box trees, the early detection and preparation of distribution maps are required. Assessment data can play an important role in this regard. The combination of high-resolution and low-spectrum panchromatic images with low resolution is used for evaluating the extent of destruction. Also, spectral and textural features are considered simultaneously in images extracted from Landsat 8 satellite. Finally, by extracting effective features from the candidate description space with the help of genetic algorithm and using the appropriate classification in the form of simultaneous application of fuzzy clustering and maximum similarity classification of area resulted in good accuracy in 2014-2018. The coefficients obtained from the models confirm their model validation for future estimates and the possibility it usage to assess the extent of the affected areas and the evolution of progress for all regions.