Contents of Volume 28 (2018)

1/2018 2/2018

3/2018

  • [1] Z. Tian, G. Wang, Y. Ren, S. Li, Y. Wang (China)
    An adaptive online sequential extreme learning machine for short-term wind speed prediction based on improved artificial bee colony algorithm , 191-212

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.012

    Abstract: As an improved algorithm of standard extreme learning machine, online sequential extreme learning machine achieves excellent classification and regression performance. However, online sequential extreme learning machine gives the same weight to the old and new training samples, and fails to highlight the importance of the new training samples. At the same time, the algorithm updates the network weights after obtaining the new training samples. This network weight updating mode lacks flexibility and increases unnecessary computation. This paper proposes an adaptive online sequential extreme learning machine with an effective sample updating mechanism. The new and old samples are given different weights. The effect of new training samples on the algorithm is further enhanced, which can further improve the regression prediction ability of extreme learning machine. At the same time, an improved artificial bee colony algorithm is proposed and used to optimize the parameters of the adaptive online sequential extreme learning machine. The stability and convergence property of proposed prediction method are proved. The actual collected short-term wind speed time series is used as the research object and verify the prediction performance of the proposed method. Multi step prediction simulation of short-term wind speed is performed out. Compared with other prediction methods, the simulation results show that the proposed approach has higher prediction accuracy and reliability performance, meanwhile improve the performance indicators.

  • [2] A. Kayabasi, A. Toktas, K. Sabanci, E. Yigit (Turkey)
    Automatic classification of agricultural grains: comparison of neural networks, 213-224

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.013

    Abstract: 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. For testing data, 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, as well. These results point out that the neural networks can be successfully applied to classification of agricultural grains whether they are properly modelled and trained.

  • [3] P. Moss, M. Svítek, M. Novák, Z. Votruba (CZ)
    Information model of resonance phenomena in brain neural networks, 225-239

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.014

    Abstract: 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 (significant) role. It is why adaptive resonance theory (ART) including resonant effects has been studied for a long time by many authors. The periodicity in the transfers of signals between the long-term memory (LTM) and short-term memory (STM) creates a possibility of resonance system structure. LTM with information content representing expectations and STM covering sensory information in resonance process offer effective learning. Nonlinear adaptive resonance creates conditions for new knowledge, or inventory observation. In the paper this feature is newly modelled by an information gyrator that best fits these linear and non-linear phenomena.

  • [4] L. Zaorálek, J. Platoš, V. Snášel (CZ)
    Patient-adapted and inter-patient ECG classification using neural network and gradient boosting , 241-254

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.015

    Abstract: Heart disease diagnosis is an important non-invasive technique. Therefore, there exists an effort to increase the accuracy of arrhythmia classification based on ECG signals. In this work, we present a novel approach of heart arrhythmia detection. The model consists of two parts. The first part extracts important features from raw ECG signal using Auto-Encoder Neural Network. Extracted features obtained by Auto-Encoder represent an input for the second part of the model, the Gradient Boosting and Feedforward Neural Network classifiers. For comparison purposes, we evaluated our approach by using MIT-BIH ECG database and also following recommendations of the Association for the Advancement of Medical Instrumentation (AAMI) for ECG class labeling. We divided our experiment into two scenarios. The first scenario represents the classification task for the patient-adapted paradigm and the second one was dedicated to the inter-patient paradigm. We compared the measured results to the state-of-the-art methods and it shows that our method outperforms the state-of-the art methods in the Ventricular Ectopic (VEB) class for both paradigms and Supraventricular Ectopic (SVEB) class in the inter-patient paradigm.

  • [5] H. Nabovati, H. Haleh, B. Vahdani (Iran)
    Fuzzy multi-objective optimization algorithms for solving multi-mode automated guided vehicles by considering machine break time and artificial neural network, 255-283

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.016

    Abstract: 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.


2/2018

  • [1] Stefanovic P., Kurasova O. (Lithunia)
    Outlier detection in self-organizing maps and their quality estimation , 105-117

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.006

    Abstract: In the paper, an algorithm that allows to detect and reject outliers in a self-organizing map (SOM) has been proposed. SOM is used for data clustering as well as dimensionality reduction and the results obtained are presented in a special graphical form. To detect outliers in SOM, a genetic algorithm-based travelling salesman approach has been applied. After outliers are detected and removed, the SOM quality has to be estimated. A measure has been proposed to evaluate the coincidence of data classes and clusters obtained in SOM. A larger value of the measure means that the distance between centers of different classes in SOM is longer and the clusters corresponding to the data classes separate better. With a view to illustrate the proposed algorithm, two datasets (numerical and textual) are used in this investigation.

  • [2] Hlavác V. (CZ)
    Genetic programming with either stochastic or deterministic constant evaluation, 119-131

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.007

    Abstract: Constant evaluation is a key problem for symbolic regression, one solved by means of genetic programming. For constant evaluation, other evolutionary methods are often used. Typical examples are some variants of genetic programming or evolutionary systems, all of which are stochastic. The article compares these methods with a deterministic approach using exponentiated gradient descent. All the methods were tested on single sample function to maintain the same conditions and results are presented in graphs. Finally, three different tasks (ten times each) are compared to check the reliability of the methods tested in the article.

  • [3] Li F., Zurada J.M., Wu W. (China,PL)
    Sparse representation learning of data by autoencoders with L1/2 regularization , 133-147

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.008

    Abstract: Autoencoder networks have been demonstrated to be effcient 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 L1/2 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 classiffcation 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 L1 regularization Autoencoder. The new representation is also illustrated to be useful for a deep network to improve the classiffcation performance.

  • [4] Huang JP., Wang XA., Zhao Y., Xin C., Xiang H. (China)
    Large earthquake magnitude prediction in Taiwan based on deep learning neural network, 149-160

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.009

    Abstract: In this paper, a deep learning-based method for earthquake prediction is proposed. Large-magnitude earthquakes and tsunamis triggered by earthquakes can kill thousands of people and cause millions of dollars worth of economic losses. The accurate prediction of large-magnitude earthquakes is a worldwide problem. In recent years, deep learning technology that can automatically extract features from mass data has been applied in image recognition, natural language processing, object recognition, etc., with great success. We explore to apply deep learning technology to earthquake prediction. We propose a deep learning method for continuous earthquake prediction using historical seismic events. First, we project the historical seismic events onto a topographic map. Taking Taiwan as an example, we generate the images of the dataset for deep learning and mark a label "1" or "0", depending on whether in the upcoming 30 days a greater than M6 earthquake will occur. Second, we train our deep leaning network model, using the images of the dataset. Finally, we make earthquake predictions, using the trained network model. The result shows that we can get the best result, when we predict earthquakes in the upcoming 30 days using data from the past 120 days. Here, we use R score as the performance metrics. The best R score is 0.303. Although the R score is not high enough, using the past 120 days' historic seismic event to predict the upcoming 30 days' biggest earthquake magnitude can be seen as the pattern of Taiwan earthquake because the R score is rather good compared to other datasets. The proposed method performs well without manually designing feature vectors, as in the traditional neural network method. This method can be applied to earthquake prediction in other seismic zones.

  • [5] Cavuslu M.A., Sahin S. (Turkey)
    FPGA implementation of ANN training using Levenberg and Marquardt algorithms , 161-178

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.010

    Abstract: Artificial Neural Network (ANN) training using gradient-based Levenberg & Marquardt (LM) algorithm has been implemented on FPGA for the solution of dynamic system identification problems within the scope of the study. In the implementation, IEEE 754 floating-point number format has been used because of the dynamism and sensitivity that it has provided. Mathematical approaches have been preferred 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 the training phase, and success rates are given for every sample set. The obtained results demonstrate that implementation of FPGA-based ANN training is possible by using LM algorithm and as the result of the training, the ANN makes a good generalization.

  • [6] Ruzek M. (CZ)
    Homeostatic learning rule for artificial neural networks , 179-189

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.011

    Abstract: This article presents an improvement of learning algorithm for an artificial neural network that makes the learning process more similar to a biological neuron, but still simple enough to be easily programmed. This idea is based on autonomous artificial neurons that are working together and at same time competing for resources; every neuron is trying to be better than the others, but also needs the feed back from other neurons. The proposed artificial neuron has similar forward signal processing as the standard perceptron; the main difference is the learning phase. The learning process is based on observing the weights of other neurons, but only in biologically plausible way, no back propagation of error or 'teacher' is allowed. The neuron is sending the signal in a forward direction into the higher layer, while the information about its function is being propagated in the opposite direction. This information does not have the form of energy, it is the observation of how the neuron's output is accepted by the others. The neurons are trying to 2nd such setting of their internal parameters that are optimal for the whole network. For this algorithm, it is necessary that the neurons are organized in layers. The tests proved the viability of this concept { the learning process is slower; but has other advantages, such as resistance against catastrophic interference or higher generalization.


1/2018

  • [1] Vareka L., Mautner P. (CZ)
    Modifications of unsupervised neural networks for single trial P300 detection , 1-16

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.001

    Abstract: P300 brain-computer interfaces (BCIs) have been gaining attention in recent years. To achieve good performance and accuracy, it is necessary to optimize both feature extraction and classification algorithms. This article aims at verifying whether supervised learning models based on self-organizing maps (SOM) or adaptive resonance theory (ART) can be useful for this task. For feature extraction, the state-of-the-art Windowed means paradigm was used. For classification, proposed classifiers were compared with state-of-the-art classifiers used in BCI research, such as Bayesian Linear Discriminant Analysis, or shrinkage LDA. Publicly available datasets from fifteen healthy subjects were used for the experiments. The results indicated that SOM-based models yield better results than ART-based models. The best performance was achieved by the LASSO model that was comparable to state-of-the-art BCI classifiers. Further possibilities for improvements are discussed.

  • [2] Horaisova K., Dudasova J., Kukal J., Rusina R., Matej R., Buncova M. (CZ)
    Discrimination between Alzheimer’s disease and amyotrophic lateral sclerosis via affine invariant spherical harmonics analysis of SPECT images , 17-39

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.002

    Abstract: Alzheimer's Disease (AD) is the most frequent form of degenerative dementia and its early diagnosis is essential for effective treatment. Functional imaging modalities including Single Photon Emission Computed Tomography (SPECT) are often used with such an aim. However, conventional evaluation of SPECT images relies on manual reorientation and visual evaluation of tomographic slices which is time consuming, subjective and therefore prone to error. Our aim is to show an automatic Computer-Aided Diagnosis (CAD) system for improving the early detection of the AD. For this purpose, ane invariant descriptors of 3D SPECT image can be useful. The method consists of four steps: evaluation of invariant descriptors obtained using spherical harmonic analysis, statistical testing of their significance, application of regularized binary index models, and model verification via leave-one-out cross-validation scheme. The second approach is based on Support Vector Machine (SVM) classifier and visualization with use of self-organizing maps. Our approaches were tested on SPECT data from 11 adult patients with definite Alzheimer's disease and 10 adult patients with Amyotrophic Lateral Sclerosis (ALS) who were used as controls. A significant difference between SPECT spherical cuts of AD group and ALS group was both visually and numerically evaluated.

  • [3] Ebadati E., Mortazavi T. (Iran)
    An efficient hybrid machine learning method for time series stock market forecasting , 41-55

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.003

    Abstract: Time series forecasting, such as stock price prediction, is one of the most important complications in the financial area as data is unsteady and has noisy variables, which are affected by many factors. This study applies a hybrid method of Genetic Algorithm (GA) and Artificial Neural Network (ANN) technique to develop a method for predicting stock price and time series. In the GA method, the output values are further fed to a developed ANN algorithm to fix errors on exact point. The analysis suggests that the GA and ANN can increase the accuracy in fewer iterations. The analysis is conducted on the 200-day main index, as well as on five companies listed on the NASDAQ. By applying the proposed method to the Apple stocks dataset, based on a hybrid model of GA and Back Propagation (BP) algorithms, the proposed method reaches to 99.99% improvement in SSE and 90.66% in time improvement, in comparison to traditional methods. These results show the performances and the speed and the accuracy of the proposed approach.

  • [4] Ramos A.D., López-Rubio E., Palomo E.J. (Spain-Equador)
    The role of the lattice dimensionality in the self-organizing map, 57-86

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.004

    Abstract: The Self-Organizing Map model considers the possibility of 1D and 3D map topologies. However, 2D maps are by far the most used in practice. Moreover, there is a lack of a theory which studies the relative merits of 1D, 2D and 3D maps. In this paper a theory of this kind is developed, which can be used to assess which topologies are better suited for vector quantization. In addition to this, a broad set of experiments is presented which includes unsupervised clustering with machine learning datasets and color image segmentation. Statistical significance tests show that the 1D maps perform significantly better in many cases, which agrees with the theoretical study. This opens the way for other applications of the less popular variants of the self-organizing map.

  • [5] Peng Z., Jiang Y., Yang X., Zhao Z., Zhang L., Wang Y. (China)
    Bus arrival time prediction based on PCA-GA-SVM, 87-104

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2018.28.005

    Abstract: Considering the correlations of the input indexes and the deficiency of calibrating kernel function parameters when support vector machine (SVM) is applied, a forecasting method based on principal component analysis-genetic algorithm-support vector machine (PCA-GA-SVM) is proposed to improve the precision of bus arrival time prediction. And the No. 232 bus in Shenyang City of China is taken as an example. The traditional SVM and Kalman Filtering model and GA-SVM are also employed to make comparative analysis on the prediction rate, respectively. The result indicates that PCA-GA-SVM obtains more accurate prediction results of bus arrival time prediction.