نشریه مهندسی برق و مهندسی کامپیوتر ایران
سال چهاردهم شماره 2 (پیاپی 39، تابستان 1395)

Particle filter is an effective tool for the object tracking problem. However, obtaining an accurate model for the system state and the observations is an essential requirement. Therefore, one of the areas of interest for the researchers is estimating the observation function according to the learning data. The observation function can be considered linear or nonlinear. The existing methods for estimating the observation function are faced some problems such as: 1) dependency to the initial value of parameters in expectation-maximization based methods and 2) requiring a set of predefined models for the multiple models based methods. In this paper, a new unsupervised method based on the kernel adaptive filters is presented to overcome the above mentioned problems. To do so, least mean squares/ recursive least squares adaptive filters are used to estimate the nonlinear observation function. Here, given the known process function and a sequence of observations, the unknown observation function is estimated. Moreover, to accelerate the algorithm and reduce the computational costs, a sparsification method based on approximate linear dependency is used. The proposed method is evaluated in two applications: time series forecasting and tracking objects in video. Results demonstrate the superiority of the proposed method compared with the existing algorithms.

With development of web-based interactions such as social networks, personal blogs, surveys and user comments, sentiment analysis and opinion mining has become an important research domain in computer science. Up to now, many approaches have been proposed for analysis of sense using machine learning and natural language processing techniques. In this paper, we used the distribution of words in the collection of documents as new criteria for analyzing sentiment. In proposed approach, we model an optimal transform domain over words distribution with two goals: maximizing spectral energy of class at low frequencies and maximizing spectral energy of at high frequencies. Using optimal transform domain, we can map data from frequency domain into Fourier domain and easily distinguish optimism and pessimism patterns. For this purpose, we use samples’ profiles of class which have low-frequency components. Assuming the contrast of the spectrum of two classes and, maximizing the spectral energy of class will be satisfied. We have performed this approach for English and Persian documents.

Holonic structures are a hierarchical formation of holons that are developed and used for the purpose of restricting interaction domains, reducing uncertainty, or forming the high level goals of multi-agent systems, in such a way that the system benefits a high degree of flexibility and dynamism in response to environmental changes.
Although the holonic multi-agent systems are extensively used in modeling and solving complex problems, most of its prerequisites, like forming the body holons and dynamically controlling its structure, use very simple application-specific models. This is due to the immaturity of the research literatures in this field. In this article, an endeavor is made to propose a goal-based approach for the formation of holonic structures, using the concepts in social science and organizational theory. The use of concepts like role, skill, and goal structures, makes the proposed method possible to be used in wide range of applications.
In order to demonstrate the capabilities of the method and also the way it can be applied in real world problems, a test bed based on the application of wireless sensor networks in object tracking is designed and presented. In this application, the sensors, which are distributed in the environment as simple agents, using holonic structures, are responsible for the track of any alien objects that enter and move in the environment. According to the empirical results of the simulations, the proposed holonic approach has provided successful performance in terms of tracking quality and energy consumption of the sensors.

Wireless senor networks (WSNs) are widely used for the monitoring purposes. One of the most challenges in designing these networks is minimizing the data transmission cost with accurate data recovery. Data aggregation using the theory of compressive sampling is an effective way to reduce the cost of communication in the sink node. The existing data aggregation methods based on compressive sampling require to a large number of nodes for each measurement sample leading to inefficient energy consumption in wireless sensor network. To solve this problem, we propose a new scheme by using sparse random measurement matrix. In this scheme, the formation of routing trees with low cost and fair distribution of load on the network significantly reduces energy consumption. Toward this goal, a new algorithm called “weighted compressive data gathering (WCDG)” is suggested in which by creating weighted routing trees and using the compressive sampling, the data belong to all of nodes of each path is aggregated and then, sent to the sink node. Considering the power control ability in sensor nodes, efficient paths are selected in this algorithm. Numerical results demonstrate the efficiency of the proposed algorithm with compared to the conventional data aggregation schemes in terms of energy consumption, load balancing, and network lifetime.

Q-learning is a one of the most popular and frequently used model-free reinforcement learning method. Among the advantages of this method is independent in its prior knowledge and there is a proof for its convergence to the optimal policy. One of the main limitations of this method is its low convergence speed, especially when the dimension is high. Accelerating convergence of this method is a challenge. Q-learning can be accelerated the convergence by the notion of opposite action. Since two Q-values are updated simultaneously at each learning step. In this paper, adaptive policy and the notion of opposite action are used to speed up the learning process by integrated approach. The methods are simulated for the grid world problem. The results demonstrate a great advance in the learning in terms of success rate, the percent of optimal states, the number of steps to goal, and average reward.

One of the key strategies to improve the efficiency is load balancing. Choosing the appropriate VM to do any task, is function of various parameters such as the amount of required resources like CPU, memory, the size of VM resource, cost and maturity of VMs. In this paper, by considering each of these criteria and design objectives such as load balancing, reducing the rate of create new VM, and VM migration, we modeling the problem in terms of effective parameters in performance. Then, we solving this model by using the PROMETHEE method, which is one of the most widely used method for MADM problems. In this method, selecting the best VM occurs based on the value assigned to each of criteria which is calculated based on fuzzy logic. To evaluate the performance of this approach, the necessary simulations have been carried out on CloudSim simulator and shown that the proposed method has better performance compared to FIFO, DLB and WRR methods on average in terms of response time, rate of success tasks, load variation and rate of VM migration.

Segmentation is necessary to determine the boundaries of the vessel. Intravascular ultrasound imaging (IVUS) is used for the diagnosis of coronary artery diseases. In this study, a new method is proposed for segmentation of IVUS images. First preprocessing is done to convert images from Cartesian coordinates to polar coordinates, remove the catheter in images and speckle noise with Nonlinear Anisotropic Diffusion Filtering. Then, texture features of an image are extracted using Gabor filter, and the image segmentation and determining the vessels boundary will be discussed using active contour without edge for vector value model. Calcium plaques have been determined using phase clustering and the exact boundary of calcium plaques is extracted using active contour model. This method has been tested on thirty images, and the results of the image segmentation have been validated by an expert. The area diffusion between the internal border and the expert’s opinion is 0.4310.236, and the area diffusion between the external border and the expert’s opinion is 0.6530.723. Area diffusion of calcium plaque extracted by the proposed algorithm compared with virtual histology images has been achieved equal to 5.90 percent.

The performance of Networks-on-Chip is highly dependent to the incorporated routing algorithms. In recent years, many routing algorithms have been proposed for 2D and 3D Networks-on-Chip. In 3D integrated circuits, different devices are stacked through silicon via in which the vertical connections are vulnerable to manufacturing process variations. Therefore, because of the high impact of faulty links or nodes on the performance of a Network-on-Chip, utilizing a fault-tolerant routing algorithm is of great importance especially for 3D Networks-on-Chip in which the vertical links are more vulnerable. In this paper, a new fault-tolerant routing algorithm called FT-ZXY is proposed to be used in 3D Networks-on-Chip. This routing method is capable of tolerating multiple vertical faulty links in addition to single horizontal faulty links without using any virtual channels thus incurs a very low hardware overhead. Experimental results reveal that the proposed routing algorithm has more reliability compared to the previous designs while incurs less latency and requires lower area and power overheads.