2018-12-12
Lukáš Embedded Svoboda
In this paper we extend Skip-Gram and Continuous Bag-of-Words models via global context information. We use Wikipedia corpus where articles are organized in a hierarchy of categories. These categories provide useful topical information about each article. We present several approaches how to enrich word meaning representation with such information. We experiment with English Wikipedia and evaluate our models on standard word similarity and word analogy datasets. Proposed models significantly outperform other word representation methods when similar size training data are used and provide similar performance compared with methods trained on much larger datasets.
2018-10-30
Mehmet Sarıgül, Mutlu Avcı
In this work, a Nadaraya-Watson kerneled learning system which owns general regression neural network topology is adapted to Q learning method to evaluate a quick and ecient action selection policy for reinforcement learning problems. By means of the proposed method Q value function is generalized and learning speed of Q agent is accelerated. The training data of the developed neural network are obtained by a standard Q learning agent on closed loop simulation system. The eciency of the proposed method is tested on populer reinforcement learning benchmarks and its performance is compared with other popular regression methods and Q-learning utilized methods.
2018-08-31
Manikandan G., Sakthi U.
Key management system maintains the confident of secret information from unauthorized users and verifying the integrity of exchanged messages and authenticity. But recent advances in electronics and computer technologies create the complexity of key management in wireless sensor networks (WSN). Additionally, the traditional key management systems are not up to the mark due to limited resources like memory, and energy constraints. In this paper, we propose an optimal cluster based key management system (OC-KMS) for WSNs. The proposed system consist of two contributions, in first, we perform the energy efficient clustering using modified animal Diaspora (MAD) optimization algorithm and cluster head (CH) selection using JAYA trust model. In second contribution, we propose the certificate less signcryption algorithm, which generates and distributes the public and private keys for each node in sensor networks. The proposed system resists various network layer attacks without affecting the network performance. The simulation result describes that the proposed system perform very efficient than existing in terms of both performance and security wise.
2018-08-30
Chang Guoqin, Hua Huo
Text sentiment analysis is an important part of social network information mining. It is also the theoretical foundation and basis of personalized recommendation, circle of interest classification and public opinion analysis. In view of the existing algorithms for feature extraction and weight calculation, we find that they fail to fully take into account the influence of emotion words. Therefore, this paper proposed a fine-grained short text sentiment analysis method based on Naive Bayes. To improve the calculation method of feature selection and weighting and proposed a more suitable sentiment analysis algorithm for features extraction named N-CHI and weight calculation named W-TF-IDF, increasing the proportion and weight of sentiment words in the feature words Through experimental analysis and comparison, the classification accuracy of this method is obviously improved compared with other methods.
2018-08-30
Yi Xu, Minghui He
The neural network model based on back-propagation (BP) algorithm is a widely used prediction model. However, the nodes number of the first hidden layer, the learning rate and momentum factor are usually determined manually, which affect the network forecast accuracy. Therefore, in this paper, to improve the forecast accuracy, firstly, the nodes number of the first hidden layer is selected adaptively based on minimizing mean square error (MSE). Secondly, improved genetic algorithm (GA) is proposed to train the learning rate and momentum factor dynamically. Thirdly, we construct a new neural network model based on the adaptively selected nodes number of the first hidden layer, the dynamically selected learning rate and momentum factor, which is called HN-GA-BP neural network model. Finally, the proposed neural network model is used to forecast the carbon dioxide levels in China for fifty years. Experimental results demonstrate the effectiveness of the proposed HN-GA-BP neural network model.
2018-08-30
Bohumír Garlik
The future of energy lies in energy self-sufficiency, renewable electricity generation, and digital technologies. The main focus is on optimizing the generation and consumption of electric power. Decentralized energy, optimization and regulation of generation and consumption of electricity to the level of the smallest producers and consumers - this is a power industry of the future. This is closely related to the solution of optimization of organisation of the electric power sources mainly from renewable energy sources (RES) in the system of their composition in the so-called RES microgrid as a decentralized power industry. A computer program will be designed to address the planning of local RES source organisation and their supply with electric power from the RES microgrid of a fictitious smart urban area. The purpose of the solution is stable energy balance in order to minimize the total cost of electricity generation, which is, in this case, determined by the prediction of its consumption in a specified period with sampling every hour. A program and a breakdown of organisation of electric power sources and their outputs covering the predicted consumption will be created. A prerequisite for the optimization or more precisely multicriterial optimization is development of special-purpose (cost, critical) function with the simultaneous definition of an optimality criterion such as penalty and formulation of restrictive conditions. Through the experiment, the conclusion of an efficient solution to the problem of optimization of electric power source organisation will be proven. The proposed optimization problem should be solved by a stochastic algorithm – simulated annealing, which is the most efficient algorithm for solution of our problem optimization. At the end, the procedures of the optimization problem solution for the special-purpose function of the RES microgrid energy system when supplying smart cities with electric power will be evaluated. The recorded annual history of electric power consumption can be sorted by day and by hour, and, then, we can see it as multidimensional data. Identification of so-called typical daily consumption patterns can be compared to a cluster analysis of the multidimensional data, where the type day charts are formed by prototypes of identified clusters correlated with a certain season. The modelled yearly history of energy consumption and the use of cluster analysis, including the use of so-called self-organizing neural network and, through the Kohenen map, the required type day charts of the relevant period have been obtained. In our experiment published hereto, we will come out of the historical electric power consumption data for working day – Wednesday. We will not address the issue of type day charts in this article; this is another very important independent scientific contribution.
2018-07-04
Zdeněk Votruba, Petr Moos, Miroslav Svítek, Mirko Novak
The paper presents an information model for representation of brain linear and nonlinear resonance phenomena based on information nullors. In the brain functions the rhythms and quasi periodicity of processes in neural networks play the outstanding role. It is why adaptive resonance theory (ART) including resonant effects is for long time studied by many authors. The periodicity in the transfers of signals between the long-term memory (LTM) and short-term memory (STM) creates possibility of resonance system structure. LTM with information content representing expectations and STM covering sensory information in resonance process offer an effective learning. Nonlinear adaptive resonance creates condition for a new knowledge, or inventory observation . In the paper this feature is newly modelled by information gyrator that best suits for these linear and non-linear phenomena.
2018-07-04
Hojat Nabovati, HASSAN HALEH, Behnam vahdani
In this paper, a novel model is presented for machines and automated guided vehicles’ simultaneous scheduling, which addresses an extension of the blocking job shop scheduling problem. An artificial neural network approach is used to estimate machine’s breakdown indexes. Since the model is strictly NP-hard and because objectives contradict each other, two developed meta-heuristic algorithms called “fuzzy multi-objective invasive weeds optimization algorithm” and “fuzzy multi-objective cuckoo search algorithm” with a new chromosome structure which guarantees the feasibility of solutions are developed to solve the proposed problem. Since there is no benchmark available on literature, three other metaheuristic algorithms are developed with a similar solution structure to validate performance of the proposed algorithms. Computational results showed that developed fuzzy multi-objective invasive weeds optimization algorithm had the best performance in terms of solving problems compared to four other algorithms.
2018-06-30
Ahmet Kayabasi, Abdurrahim Toktas, Kadir Sabanci, Enes Yigit
In this study, applications of well-known neural networks such as artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS) and support vector machine (SVM) for wheat grain classification into three species are comparatively presented. The species of wheat grains which are Kama (#70), Rosa (#70) and Canadian (#70) are designated as outputs of neural network models. The classification is carried out through data of wheat grains (#210) acquired using X-ray technique. The data set includes seven grain’s geometric parameters: Area, perimeter, compactness, length, width, asymmetry coefficient and groove length. The neural networks input with the geometric parameters are trained through 189 wheat grain data and their accuracies are tested via 21 data. The performance of neural network models is compared to each other with regard to their accuracy, efficiency and convenience. The ANN, ANFIS and SVM models numerically calculate the outputs with mean absolute error (MAE) of 0.014, 0.018 and 0.135, and classify the grains with accuracy of 100%, 100% and 95.23%, respectively. Furthermore, data of 210 grains is synthetically increased to 3210 in order to investigate the proposed models under big data. It is seen that the models are more successful if the size of data is increased. These results point out that the neural networks can be successfully applied to classification of agricultural grains whether they are properly modelled and trained.
2018-04-30
Petr Berka
Credit risk assessment, credit scoring or loan applications approval are one of the typical tasks that can be solved using machine learning or data mining techniques. From this perspective, loan applications evaluation is a classification task, in which the final decision can be either a crisp yes/no decision about the loan or a numeric score expressing the financial standing of the applicant. The knowledge to be used is inferred from data about past decisions. These data usually consists off socio-demographic characteristics, economic characteristics (e.g. income, deposit), the characteristics of the loan, and, the loan approval decision. A number of machine learning algorithms can be used for this purpose. In this paper we show how this task can be solved using the LISp-Miner system, a tool that is under development at the University of Economics, Prague. LISp-Miner is primary oriented on mining for various types of association rules, but unlike "classical" association rules proposed by Agrawal, LISp-Miner introduces a greater variety of different types of relations between left-hand and right-hand side of a rule. Beside this, two other procedures that can be used for classification task are implemented in LISp-Miner as well. We describe the 4FT-Miner and KEX procedures and show how they can be used to analyze data about loan applications. We also compare the results obtained using the presented algorithms with results from standard rule learning methods.
2018-04-30
Jana Kuklová, Ondřej Přibyl
This paper deals with the changeover from decision tree (bivalent logic) approach to fuzzy logic approach to highway traffic control, particularly to variable speed limits display. The usage of existing knowledge from decision tree control is one of the most suitable methods for identication of new fuzzy model. However, such method introduces several difficulties. These difficulties are described and possible measures are proposed. Several fuzzy logic algorithms were developed and tested by a microsimulation model. The results are presented and the finest algorithm is recommended for testing on the Prague City Ring Road in real conditions. This paper provides a guidance for researchers and practitioners dealing with similar problem formulation.
2018-04-27
Mehmet Ali Çavuşlu, Suhap Şahin
ANN training using gradient based LM algorithm has been implemented on FPGA for the solution of dynamic system identification problems within the scope of study. In the implementation, IEEE 754 floating-point number format has been used because of dynamism and sensitivity that it has provided. Mathematical approaches have been preferred in order to implement the activation function, which is the most critical phase of the study. ANN is tested by using input-output sample sets, which are shown or not shown to the network in training phase, and success rates are given for every sample set. The results obtained show that it is possible to iplementation of ANN training as FPGA based by using LM algorithm and that the ANN make a good generalization.
2018-04-27
Feng Li, Jacek M. Zurada, Wei Wu
Autoencoder networks have been demonstrated to be efficient for unsupervised learning of representation of images, documents and time series. Sparse representation can improve the interpretability of the input data and the generalization of a model by eliminating redundant features and extracting the latent structure of data. In this paper, we use L_1/2 regularization method to enforce sparsity on the hidden representation of an autoencoder for achieving sparse representation of data. The performance of our approach in terms of unsupervised feature learning and supervised classification is assessed on the MNIST digit data set, the ORL face database and the Reuters-21578 text corpus. The results demonstrate that the proposed autoencoder can produce sparser representation and better reconstruction performance than the Sparse Autoencoder and the L_1 regularization Autoencoder. The new representation is also illustrated to be useful for a deep network to improve the classification performance.