Portfolio optimization under classic mean-variance framework of Markowitz must be revised as variance fails to be a good risk measure. This is especially true when the asset returns are not normal. In this paper, we utilize Value at Risk (VaR) as the risk measure and Historical Simulation (HS) is used to obtain an acceptable estimate of the VaR. Also, a well known multi-objective evolutionary approach is used to address the inherent bi-objective problem; In fact, NSGA-II is incorporated here. This method is tested on a set of past return data of 12 assets on Tehran Stock Exchange (TSE). A comparison of the obtained results, shows that the proposed method offers high quality solutions and a wide range of risk return trade-offs.

One of the most effective ways to extract knowledge from large information resources is applying data mining methods. Since the amount of information on the Internet is exploding, using XML documents is common as they have many advantages. Knowledge extraction from XML documents is a way to provide more utilizable results. XCLS is one of the most efficient algorithms for XML documents clustering. In this paper we represent a new algorithm for clustering XML documents. This algorithm is an improvement over XCLS algorithm which tries to obviate its problems. We implemented both algorithms and evaluated their clustering quality and running time on the same data sets. In both cases, it is shown that the performance of the new algorithm is better.

Recently, a technique has been proposed for image hiding, that is based on block texture similarity where, blocks of secret image are compared with blocks of a set of cover images and the cover image with the most similar blocks to those of the secret image is selected as the best candidate cover image to conceal the secret image. In this paper, we propose a new image hiding method in which, the secret image is initially coded using a fuzzy coding/decoding method. By applying the fuzzy coder, each block of the secret image is compressed to a smaller block. In this way, after compressing the secret image to a smaller one, we hide it in a cover image. Obviously hiding a smaller secret image causes less distortion in the stego-image (the image that has secret image or data) and therefore higher quality stego-image is obtained. Consequently, the proposed method provides higher embedding rate and enhanced security.

Two popular and better performing approaches to language Identification (LID) are Phone Recognition followed by Language Modeling (PRLM) and Parallel PRLM. In this paper, we report several improvements in Phone Recognition which reduces error rate in PRLM and PPRLM based LID systems. In our previous paper, we introduced APRLM approach that reduces error rate for about 1.3% in LID tasks. In this paper, we suggest other solution that overcomes APRLM. This new LID approach is named Generalized PRLM or GPRLM. Several language identification experiments were conducted and the proposed improvements were evaluated using OGI-MLTS corpus. Our results show that GPRLM overcomes PPRLM and APRLM about 2.5% and 1.2% respectively in two language classification tasks.

This paper introduces a revisited hybrid algorithm for function approximation. In this paper, a simple and fast learning algorithm is proposed, which automates structure and parameter identification simultaneously based on input-target samples. First, without need of clustering, the initial structure of the network with the specified number of rules is established, and then a training process based on the error of other training samples is applied to obtain a more precision model. After the network structure is identified, an optimization learning, based on the criteria error, is performed to optimize the obtained parameter set of the premise parts and the consequent parts. At the end, comprehensive comparisons are made with other approaches to demonstrate that the proposed algorithm is superior in term of compact structure, convergence speed, memory usage and learning efficiency.

This paper presents a new segmentation method for color images. It relies on soft and hard segmentation processes. In the soft segmentation process, a cellular learning automata analyzes the input image and closes together the pixels that are enclosed in each region to generate a soft segmented image. Adjacency and texture information are encountered in the soft segmentation stage. Soft segmented image is then fed to the hard segmentation process to generate the final segmentation result. As the proposed method is based on CLA it can adapt to its environment after some iterations. This adaptive behavior leads to a semi content-based segmentation process that performs well even in presence of noise. Experimental results show the effectiveness of the proposed segmentation method.

In this paper, a novel inference engine named fuzzyevidential hybrid engine has been proposed using Dempster-Shafer theory of evidence and fuzzy sets theory. This hybrid engine operates in two phases. In the first phase, it models the input information’s vagueness through fuzzy sets. In following, extracting the fuzzy rule set for the problem, it applies the fuzzy inference rules on the acquired fuzzy sets to produce the first phase results. At second phase, the acquired results of previous stage are assumed as basic beliefs for the problem propositions and in this way, the belief and plausibility functions (or the belief interval) are set. Gathering information from different sources, they provide us with diverse basic beliefs which should be fused to produce an integrative result. For this purpose, evidential combination rules are used to perform the information fusion. Having applied the proposed engine on the coronary heart disease (CHD) risk assessment, it has yielded 86 percent accuracy rate in the CHD risk prediction.

One of the toughest challenges in medical diagnosis is uncertainty handling. The recognition of intestinal bacteria such as Salmonella and Shigella which cause typhoid fever and dysentery, respectively, is one such challenging problem for microbiologists. In this paper, we take an intelligent approach towards the bacteria classification problem by using five similarity measures of fuzzy sets (FSs) and intuitionistic fuzzy sets (IFSs) to examine their capabilities in encountering uncertainty in the medical pattern recognition. Finally, the recognition rates of the measures are calculated among which IFS Mitchel and Hausdorf similarity measures score the best results with 95.27% and 94.48% recognition rates, respectively. On the other hand, FS Euclidean distance yieldes only 85% recognition rate.

PSO, like many stochastic search methods, is very sensitive to efficient parameter setting. As modifying a single parameter may result in a large effect. In this paper, we propose a new a new learning automatabased approach for adaptive PSO parameter selection. In this approach three learning automata are utilized to determine values of each parameter for updating particles velocity namely inertia weight, cognitive and social components. Experimental results show that the proposed algorithms compared to other schemes such as SPSO, PSO-IW, PSO TVAC, PSO-LP, DAPSO, GPSO, and DCPSO have the same or even higher ability to find better local minima. In addition, proposed algorithms converge to stopping criteria significantly faster than most of the PSO algorithms.

In real world, optimization problems are usually dynamic in which local optima of the problem change. Hence, in these optimization problems goal is not only to find global optimum but also to track its changes. In this paper, we propose a variant of cellular PSO, a new hybrid model of particle swarm optimization and cellular automata, which addresses dynamic optimization. In the proposed model, population is split among cells of cellular automata embedded in the search space. Each cell of cellular automata can contain a specified number of particles in order to keep the diversity of swarm. Moreover, we utilize the exploration capability of quantum particles in order to find position of new local optima quickly. To do so, after a change in environment is detected, some of the particles in the cell change their role from standard particles to quantum for few iterations. Experimental results on moving peaks benchmark show that the proposed algorithm outperforms mQSO, a well-known multi swarm model for dynamic optimization, in many environments.

Particle Swarm Optimization (PSO) is an optimization method that is inspired by nature and is used frequently nowadays. In this paper we proposed a new dynamic geometric neighborhood based on Voronoi diagram in PSO. Voronoi diagram is a geometric naturalistic method to determine neighbors in a set of particles. It seems that in realistic swarm, particles take Voronoi neighbors into account. Also a comparison is made between the performance of some traditional methods for choosing neighbors and new dynamic geometric methods like Voronoi and dynamic Euclidean. In this comparison it is found that PSO with geometric neighborhood can achieve better accuracy overall especially when the optimum value is out of the initial range.

In recent years, processing the images that contain human faces has been a growing research interest because of establishment and development of automatic methods especially in security applications, compression, and perceptual user interface. In this paper, a new method has been proposed for multiple face detection and tracking in video frames. The proposed method uses skin color, edge and shape information, face detection, and dynamic movement analysis of faces for more accurate real-time multiple face detection and tracking purposes. One of the main advantages of the proposed method is its robustness against usual challenges in face tracking such as scaling, rotation, scene changes, fast movements, and partial occlusions.

In this paper, we present a new measure for evaluating similarity changes in a multi agent system. The similarity measure of the agents changes during the learning process. The similarity differences are because of any composition or decomposition of some agent sets. The presented measure, defines the changes of homogeneity of agents by composition and decomposition. The utility of the metrics is demonstrated in the experimental evaluation of multi agent foraging. The results show that while the similarity difference gets a positive value, the performance grow rapidly.

In this paper we propose several methods for partitioning, the process of grouping members of population to different memeplexes, in a shuffled frog leaping algorithm. These proposed methods divide the population in terms of the value of cost function or the geometric position of members or quite random partitioning. The proposed methods are evaluated on several low and high dimensional benchmark functions. The obtained results on low dimensional functions demonstrate that geometric partitioning methods have the best success rate and the fastest performance. Also on high dimensional functions, however using of the geometric partitioning methods for the partitioning stage of the SFL algorithm lead to a better success rate but these methods are more time consuming than other partitioning methods.

Keystroke dynamics-based authentication, KDA, verifies users via their typing patterns. To authenticate users based on their typing samples, it is required to find out the resemblance of a typing sample and the training samples of a user regardless of the text typed. In this paper, a measure is proposed to find the distance between a typing sample and a set of samples of a user. For each digraph, histogram-based density estimation is used to find the pdf of its duration time. This measure is combined with another measure which is based on the two samples distances. Experimental results show considerable decrease in FAR while FRR remains constant.

Increasing the domain of locality by using treeadjoining- grammars (TAG) encourages some researchers to use it as a modeling formalism in their language application. But parsing with a rich grammar like TAG faces two main obstacles: low parsing speed and a lot of ambiguous syntactical parses. We uses an idea of the shallow parsing based on a statistical approach in TAG formalism, named supertagging, which enhanced the standard POS tags in order to employ the syntactical information about the sentence. In this paper, an error-driven method in order to approaching a full parse from the partial parses based on TAG formalism is presented. These partial parses are basically resulted from supertagger which is followed by a simple heuristic based light parser named light weight dependency analyzer (LDA). Like other error driven methods, the process of generation the deep parses can be divided into two different phases: error detection and error correction, which in each phase, different completion heuristics applied on the partial parses. The experiments on Penn Treebank show considerable improvements in the parsing time and disambiguation process.

A watermarking method in DCT domain is modified to achieve better imperceptibility. Particle Swarm Optimization (PSO) is used to find the best DCT coefficients for embedding the watermark sequence and the Structural Similarity Index is used as the fitness function in order to have a watermarked image with the best possible quality.

In a quest for modeling human brain, we are going to introduce a brain model based on a general framework for brain called Memory-Prediction Framework. The model is a hierarchical Bayesian structure that uses Reservoir Computing methods as the state-of-the-art and the most biological plausible Temporal Sequence Processing method for online and unsupervised learning. So, the model is called Hierarchical Bayesian Reservoir Memory (HBRM). HBRM uses a simple stochastic gradient descent learning algorithm to learn and organize common multi-scale spatio-temporal patterns/features of the input signals in a hierarchical structure in an unsupervised manner to provide robust and real-time prediction of future inputs. We suggest HBRM as a real-time high-dimensional stream processing model for the basic brain computations. In this paper we will describe the model and assess its prediction accuracy in a simulated real-world environment.

This paper describes the theory and implementation of dynamic Bayesian networks in the context of speaker identification. Dynamic Bayesian networks provide a succinct and expressive graphical language for factoring joint probability distributions, and we begin by presenting the structures that are appropriate for doing speaker identification in clean and noisy environments. This approach is notable because it expresses an identification system using only the concepts of random variables and conditional probabilities. We present illustrative experiments in both clean and noisy environments and our experiments show that this new approach is very promising in the field of speaker identification.

A new structure learning approach for Bayesian networks (BNs) based on asexual reproduction optimization (ARO) is proposed in this paper. ARO can be essentially considered as an evolutionary based algorithm that mathematically models the budding mechanism of asexual reproduction. In ARO, a parent produces a bud through a reproduction operator; thereafter the parent and its bud compete to survive according to a performance index obtained from the underlying objective function of the optimization problem; this leads to the fitter individual. The proposed method is applied to real-world and benchmark applications, while its effectiveness is demonstrated through computer simulation. Results of simulation show that ARO outperforms GA because ARO results good structure in comparison with GA and the speed of convergence in ARO is more than GA. Finally, the ARO performance is statistically shown.