Search

The aim of this article is to introduce a hybrid approach, namely optimal multiple kernel–support vector regression (OMK–SVR) for time series data prediction and to analyze and compare its performances against those of support vector regression with a single RBF kernel (RBF-SVR), gene expression programming (GEP) and extreme learning machine (ELM) on the financial series formed by the monthly and weekly values of Bursa Malaysia KLCI Index, monthly values of Dow Jones Industrial Average Index (DJIA) and New York Stock Exchange. Our method provides an optimal multiple kernel and optimal parameters in Support Vector Regression, improving the accuracy of prediction. The proposed approach is structured on two levels. The macro-level uses a breeder genetic algorithm for choosing the optimal multiple kernel and the SVR optimal parameters. The fitness function of each chromosome is computed in the micro-level using a SVR algorithm. The regression model based on the optimal multiple kernel and optimal parameters is then validated and used for forecasting. The experimental results prove that OMK–SVR performs better than GEP, RBF-SVR and ELM for predicting the future behavior of the study series. A sensitivity study with respect to the number of kernels from the multiple kernel used by OMK–SVR and with respect to the ratio between training and testing data sets was conducted.

The goal of this article is to propose a framework for automatic identification of watermarks from modified host images. The framework can be used with any watermark embedding/extraction system and is based on models built using machine learning (ML) techniques. Any supervised ML approach can be theoretically chosen. An important part of our framework consists in building a stand-alone module, independent of the watermarking system, for generating two types of watermarks datasets. The first type of datasets, that we will name artificially datasets, is generated from the original images by adding noise with an imposed maximum level of noise. The second type contains altered watermarked images obtained from the original ones by using different transformations. The module also performs an automatic labeling process of these data, building watermarks’ containers. Then, many models can be built using the watermarks containers and different ML techniques. Comparing the performances of all the obtained models allows the choice of the best model, or provides details for building ensemble learning. To validate the proposed framework, we conducted experiments using a particular watermarking system, built by us and many models based on artificial neural networks (ANN) and support vector machines (SVM). As a side result we identified a possible methodology for evaluating the robustness of a watermarking system, by using ANN and the two types of datasets generated in our proposed methodology.

This paper contributes to the design of realistic autonomous agents which can simulate the human behavior and respond to multiple factors such as impulse, emotions, social influences, etc.

The General Data Protection Regulation (GDPR) has entered into force in the European Union (EU) since 25 May 2018 in order to satisfy present difficulties related to private information protection. This regulation involves significant structural for companies, but also stricter requirements for personal data collection, management, and protection. In this context, companies need to create smart solutions to allow them to comply with the GDPR and build a feeling of confidence in order to map all their personal data. In these conditions, cognitive computing could be able to assist companies extract, protect and anonymize sensitive structured and unstructured data. Therefore, this article proposes a framework that can serve as an approach or guidance for companies that use cognitive computing methods to meet GDPR requirements. The goal of this work is to examine the smart system as a data processing and data protection solution to contribute to GDPR compliance.

Clustering is an important task in machine learning and data mining. Due to various applications that use clustering, numerous clustering methods were proposed. One well-known, simple, and widely used clustering algorithm is k-means. The main problem of this algorithm is its tendency of getting trapped into local minimum because it does not have any kind of global search. Clustering is a hard optimization problem, and swarm intelligence stochastic optimization algorithms are proved to be successful for such tasks. In this paper, we propose recent swarm intelligence elephant herding optimization algorithm for data clustering. Local search of the elephant herding optimization algorithm was improved by k-means. The proposed method was tested on six benchmark datasets and compared to other methods from literature. Based on the obtained results it can be concluded that the proposed method finds better clusters when silhouette score is used as the quality measure.

Single hidden layer feed forward neural networks are widely used for various practical problems. However, the training process for determining synaptic weights of such neural networks can be computationally very expensive. In this paper we propose a new learning algorithm for learning the synaptic weights of the single hidden layer feedforward neural networks in order to reduce the learning time. We propose combining the upgraded bat algorithm with the extreme learning machine. The proposed approach reduces the number of evaluations needed to train a neural network and efficiently finds optimal input weights and the hidden biases. The proposed algorithm was tested on standard benchmark classification problems and functions and compared with other approaches from literature. The results have shown that our approach produces a satisfactory performance in almost all cases and that it can obtains solutions much faster than the traditional learning algorithms.

The aim of this paper is to propose a model for an adaptive multi-agent system based on wasp-like behaviour for dynamic allocation of puzzles and quests in the virtual learning game SOTIRIOS. This is a digital learning game integrated inside a First Person Shooter designed by the second author of this paper. The learning process is based on many puzzles hidden in the game flow. The multi-agent system is necessary to integrate a multiplayer mode into the game. The agents use wasp task allocation behaviour, combined with a model of wasp dominance hierarchy in order to create a unique multiplayer learning system, where each user has a different learning curve, based on his results. The wasp behaviour is required to create a balanced multiplayer mode and to optimize the results of teams within the game.

For most wireless sensor networks applications it is necessary to know the locations of all sensor nodes. Since sensor nodes are usually cheap, it is impossible to equip them all with GPS devices, hence the localization process depends on few static or mobile anchor nodes with GPS devices. Range based localization methods use estimated distance between sensor and anchor nodes where the quality of estimation usually depends on the distance and angle of arrival. Localization based on such noisy data represents a hard optimization problem for which swarm intelligence algorithms have been successfully used. In this paper we propose a range based localization algorithm that uses recently developed bat algorithm. The two stage localization algorithm uses four semi-mobile anchors that are at first located at the corners of the area where sensors are deployed and after that the anchors move to their optimal positions with minimal distances to sensor nodes, but with maximal viewing angles. Our proposed algorithm is even at the first stage superior to other approaches from literature in minimizing the error between real and estimated sensor node positions and it is additionally improved at the second stage.

The aim of this paper is to present and compare some approaches for obtaining multivariate interpolation spaces. The comparison is made using a set of conditions of Hermite–Birkhoff type. Algorithms for these approaches are also presented. We prove that the interpolation space for these conditions is the set of polynomials of degree n in 2 variables.

Most of our knowledge regarding the process of protein import into mitochondria has come from research employing Saccharomyces cerevisiae as a model system. Recently, several mammalian homologues of the mitochondrial motor proteins were identified. Of particular interest for us is the human Tim14/Pam18-Tim16/Pam16 complex. We chose a structural approach in order to examine the evolutionary conservation between yeast Tim14/Pam18-Tim16/Pam16 proteins and their human homologues. For this purpose, we examined the structural properties of the purified human proteins and their interaction with their yeast homologues, in vitro. Our results show that the soluble domains of the human Tim14/Pam18 and Tim16/Pam16 proteins interact with their yeast counterparts, forming heterodimeric complexes and that these complexes interact with yeast mtHsp70.

The aim of this paper is to define a general frame for building optimal multiple SVM kernels. Our scheme follows 5 steps: formal representation of the multiple kernels, structural representation, choice of genetic algorithm, SVM algorithm, and model evaluation. The computation of the optimal parameter values of SVM kernels is performed using an evolutionary method based on the SVM algorithm for evaluation of the quality of chromosomes. After the multiple kernel is found by the genetic algorithm we apply cross validation method for estimating the performance of our predictive model. We implemented and compared many hybrid methods derived from this scheme. Improved co-mutation operators are used and a comparative study about their effect on the predictive model performances is made. We tested our multiple kernels for classification tasks but they can be also used for other types of tasks.

The aim of this paper is to present a new method for optimization of SVM multiple kernels The kernel substitution can be used to define many other types of learning machines distinct from SVMs We introduced a new hybrid method which uses in the first level an evolutionary algorithm based on wasp behaviour and on the co-mutation operator LR−Mijn and in the second level a SVM algorithm which computes the quality of chromosomes The most important details of our algorithms are presented The testing and validation proves that multiple kernels obtained using our genetic approach are improving the classification accuracy up to 94.12% for the “leukemia” data set.

Multivariate interpolation is a topic which often appears in practical modeling problems. Different type of spaces of functions are used for solving interpolation problems. When the interpolation conditions are of different kind, by example, spacial and temporal, one possibility for modeling the problem is to use a generalize degree, in which the monomials exponents are weighted with a weight vector with integer components. In order to use such a generalize polynomial space as interpolation space, it is necessary to know the dimension and a basis of it. The aim of this article is to study and prove many properties of the generalize polynomial spaces in three variables.

Broken packages are packages with not enough information regarding the destination address and they have a negative impact on a computer network. Broken packages stay in the router too much time and overload it. This is a real problem which appears in the computer network of Sofia University. We know only the time when a broken package arrives in the router. The problem is to find the computer which send broken packages. The aim of this article is to propose a model for defining the data transmission stability and correctness in a computer network. Our theoretical model is temporal dependent and captures the transmission properties of a computer network.

The aim of this paper is to illustrate a possibility of obtaining various theoretical results using the connection between multivariate interpolation and reduction process with respect to a H-basis of an ideal. Using this connection we can switch between interpolation theory and the theory of ideals. As a application of this connection, we found and proved an interesting identity, which is satisfied for all polynomials in d variables from an interpolation polynomial subspace.

In this paper is presented a model for an adaptive multi-agent system for dynamic routing of the grants' activities from a learning environment, based on the adaptive wasp colonies behavior. The agents use wasp task allocation behavior, combined with a model of wasp dominance hierarchy formation. The model we introduced allows the assignment of activities in a grant, taking into account the specialization of students, their experience and the complexity of activities already taken. An adaptive method allows students to enter in the Grant system for the first time. The system is changing dynamic, because both the type of activities and the students involved in the system change. Our approach depends on many system's parameters. For the implementation these parameters were tuned by hand. The Grant-system we built is integrated in a virtual education system, student centered, that facilitates the learning through collaboration as a form of social interaction.

We have designed a virtual learning environment where students interact through their computers and with the software agents in order to achieve a common educational goal. The Multi-Agent System (MAS) consisting of autonomous, cognitive and social agents communicating by messages is used to provide a group decision support system for the learning environment. Learning objects are distributed in a network and have different weights in function of their relevance to a specific educational goal. The relevance of a learning object can change in time; it is affected by students', agents' and teachers' evaluation. We have used an ant colony behavior model for the agents that play the role of a tutor and organizing the group-work activities for the students.

The aim of this article is to obtain a quadrature formula for functions in several variables and to analyze the algorithmic and computational aspects of this formula. The known information about the integrand is {λi(f)}i=1n, where λi are linearly independent linear functionals. We find a form of the coefficients of the quadrature formula which can be easy used in numerical calculations. The main algorithm we use in order to obtain the coefficients and the remainder of the quadrature formula is based on the Gauss elimination by segments method. We obtain an expression for the exactness degree of the quadrature formula. Finally, we analyze some computational aspects of the algorithm in the particular case of the Lagrange conditions.

Multivariate ideal interpolation schemes are deeply connected with H-bases. Both the definition of a H-basis and of an ideal interpolation space depend of the notion of degree used in the grading decomposition of the polynomial spaces. We studied, in the case of bivariate polynomials, a generalized degree, introduced by T. Sauer and named w-degree. This article give some theoretical results that allow us to construct algorithms for calculus of the dimension of the homogeneous spaces of bivariate polynomials of w – degree n. We implemented these algorithms in C++ language. The analysis of the results obtained, leads us to another theoretical conjecture which we proved in the end.

The aim of this article is to introduce and to study a generalization of the error order of interpolation, named λ – error order of interpolation. This generalization makes possible a deeper analysis of the error in the interpolation process. We derived the general form of the λ – error order of interpolation and then we applied it for many choices of the functional λ.