نشریه محاسبات نرم
سال سوم شماره 2 (پیاپی 6، پاییز و زمستان 1393)

Determining the optimum number of nodes، number of hidden layers، and synaptic connection weights in an artificial neural network (ANN) plays an important role in the performance of this soft computing model. Several methods have been proposed for weights update (training) and structure selection of the ANNs. For example، the error back-propagation (EBP) is a traditional method for weights update of multi-layer networks. In this study، gravitational search algorithm (GSA)، as a modern swarm intelligence optimization method، is used for this purpose. In this way، GSA and its binary version (BGSA) are used concurrently، as a hybrid method، to optimize the weights and number of hidden-layer nodes of an ANN، respectively. The performance of proposed method is compared with other intelligent and traditional methods such as particle swarm optimization (PSO)، PSO-BPSO hybrid، and EBP in classification and function approximation tasks. The performance is evaluated over Iris، Breast Cancer، and Glass datasets for classification. For function approximation task، the performance is evaluated for a prosody predictor that is used for natural speech synthesis. To reduce the number of inputs to prosody predictor، a hybrid of an genetic algorithm and ant colony optimization algorithm is used for feature selection. Simulation results show that the proposed method can offer competitive classification and prediction accuracies while using a reduced number of hidden neurons (25-68 percent reduction) as compared to other investigated algorithms.

In this paper، a new approach for selective harmonic elimination (SHE) in a cascaded multilevel inverter is proposed. The switching angles are determined with the assumption of varying input DC sources and at this condition the fundamental component is remained adjusted and undesired harmonic components are eliminated. The on-line switching angles determination is done by an Artificial Neural Network (ANN). The training data set for ANN are produced by application of heuristic algorithms to solve the SHE problem. In this paper، various algorithms such as harmony search algorithm، imperialist competitive algorithm and particle swarm optimization are applied to obtain the switching angles and by comparing the results of these methods، the better one is used for ANN training.

Due to enormous volume of information available on the Web، finding appropriate knowledge in a short time seems difficult. Knowledge Recommender systems، Online Forums and Question Answering (QA) systems were created to facilitate finding suitable information. QA systems use knowledge repositories to retrieve brief responses to users’ queries. Expert Finding system، not only causes knowledge transition from responder to requester، but also it improves requester’s experience and thoughts about the subject. These systems use textual analysis، historical activities analysis، social relation analysis and combination of these approaches to find experts. In this paper، we used resume data of university professors to develop a new model which outperforms conventional methods by using Vector Space Model and Query Expansion Technique.

The growing pace of cancer has encouraged researchers to deliberate several aspects of this malignant disease. Genetic-induced nature of cancer، heighten the importance of studying intra-cell components. This paper has been carried out with the aim of making some specific and unique features clear from those long DNA sequences by employing well-established DNA sequence analysis techniques. The identical part of human body DNA sequences have been used to simulate proposing algorithm. Z-Curve mapping method has been utilized in order to conversion of DNA alphabetic strings to digital signals. This method has made use of Linear Predictive Coding (LPC) model to analyze resultant data for feature extraction. In addition، this paper is beneficiary of a certain singular value decomposition (SVD) computational approaches to select significant features for dimension reduction. Finally، statistical parameters discriminate cancerous samples from non-cancerous ones. This discrimination represents the biological mutation concept which expresses the genetic changes of cancer disease.

One of the main challenges in Wireless Sensor Networks is the limited energy of nodes which can cause to reduce the lifetime of nodes and whole network respectively. Transmissions between the nodes consumes most of the nodes'' energy so minimization of unnecessary transmissions can led to reduction of energy consumption. Therefor routing protocols designed based on optimal energy consumption are necessary. When the sensor nodes are deployed densely، data sensed by the nodes may، to some extent، be the same or similar and so are unnecessary. Energy dissipation and high traffic are costs that must be paid for transmission of repeated data from the source nodes to the base station. In order to solve these issues and optimize the energy consumption in communication، information gathering is considered as an effective technique. Redundancy in the raw data can be deleted in the relay nodes. The only useful information is retrieved and sent to the base station. Therefore، the number of data packets received at the base station، will be reduced، and thus save energy consumption and bandwidth. Optimal gathering of data can be determined in term of total energy consumed to transfer data from nodes to the base station. Conventional routing protocols which use data collection techniques، can be divided into two categories based on network architecture: tree protocol and clustering protocols. In this study، we focus on the second type of network architecture، the tree protocol. In these protocols، sensor nodes are organized in a tree and aggregation is done at intermediate nodes in the junction tree branches. Gathered data packets are then transmitted to the root which is the base station. Tree protocols that are suitable for applications dealing with the aggregation of data within the network. One of the main characteristics of tree protocols is optimizing the data aggregation tree structure based on energy consumption which we do this benefiting a swarm intelligence algorithm called River Formation Dynamics. Simulation results show that our proposed algorithm will give better results in terms of network lifetime compared with ant colony algorithm.

Accurate segmentation plays a vital role in automated analysis of ultrasonic images. A new method based on adaptive anisotropic diffusion is introduced here for lesion detection in ultrasonic images of the breast. In this method، a hypothesis testing framework is defined first to separate lesions from healthy breast tissue. Then the boundary of lesion is estimated by adaptive anisotropic diffusion followed by energy minimization method. The performance of the proposed method is evaluated on real ultrasound images which are captured from various breast tissues containing different lesions. Parameters for performance evaluation are presented and obtained results are compared with some other state-of-the-art lesion detection methods. Increased true positive rate in parallel with simultaneous decrease in false positive rate، Hammoude distance and Hausdorff distance confirm the effectiveness of proposed method for detecting breast cancer.