International Journal of Machine Intelligence and Sensory Signal Processing (5 papers in press)
Optimal selection of learning parameters for regularized random vector functional-link networks-based soft measuring model
by Jian Tang
Abstract: Random vector functional-link networks (RVFLN) with one single hidden layer structure have been used widely for soft measuring model construction. In which, the input weights and biases are produced randomly, and the output weights are computed analytically by a Moore-Penrose generalized inverse method. Regularized RVFLN (RRVFLN) can prevent over-fitting problem and reduce complexity of the constructed model by using the ridge regression method. Several learning parameters, such as range of random input weights and bias, number of hidden nodes and regularizing factor, are data dependent. This paper aims to develop a composite differential evolution (CoDE)-based optimal selection method to address the three learning parameters of RRVFLN. Experiments on some benchmark datasets are carried out to validate the proposed method.
Keywords: Random vector functional-link networks (RVFLN); Learning parameters optimal selection; Range of random input weights and bias; Composite differential evolution (CoDE).
Time-Aware Efficient Prediction and Anomaly Detection for Large-Scale Light Curves
by Jing Bi, Tianzhi Feng, Haitao Yuan
Abstract: In the era of data explosion, how to process large-scale data is one of the most important problems. In the field of astronomy, the growth of data is particularly rapid. This paper focuses on the processing of large-scale astronomical data. Stellar brightness is an important attribute of the stars. Ground-based Wide-Angle Camera array (GWAC) can provide a huge volume of data for the brightness analysis of numerous stars. Based on the GWAC data, this work aims to analyze and predict the light curves, as well as to conduct early detection of the abnormal variation in brightness of stars for the special astro- nomical phenomena. A Parallel Auto-Regressive Integrated Moving Average (PARIMA) model is proposed for the time series analysis. The data are analyzed and predicted by the acquisition and processing of Mini-GWAC data. To reduce the data processing time, this paper incorporates the multi-process mechanism into the ARIMA model to process the mini-GWAC data. After determining the parameters, the model is used to predict the abnormal phenomena. This work can predict each single point five times and dynamically set early warning limits to make the alarm more accurate. The alarm decision can be determined based on the mean value of five prediction errors and early warning limits. Thus, the proposed PARIMA method can accurately predict and alarm in time.
Keywords: PARIMA; real-time analysis; big data processing; light curve; anomaly detection.
Review of techniques for predicting hard drive failure with SMART attributes
by Marco Garcia, Vladimir Ivanov, Anastasia Kozar, Stanislav Litvinov, Alexey Reznik, Vitaly Romanov, Giancarlo Succi
Abstract: Hard drive failure prediction is still a relevant problem today. Since the introduction of SMART, which aimed at providing warnings about failure, a number of statistical and machine learning techniques were proposed to improve warnings accuracy. Failure prediction is a part of reliability analysis - the field that was studied intensively for decades. However, availability of techniques for failure prognosis does not automatically imply their applicability for a specific case. SMART was developed to provide a meaningful information that can signify the health status of a hard drive. Failure prognosis techniques can leverage this information to provide timely and reliable warnings. In this paper, we review existing datasets with SMART attributes, feature selection methods from the set of SMART attributes, and hard drive failure prediction techniques based on SMART, in an attempt to find the best algorithm and evaluate its applicability for regular use.rn
Keywords: Reliability; failure modelling; cyberphysical systems; machine intelligence.
Multi-view data ensemble clustering: A cluster-level perspective
by Jiye Liang, Qianyu Shi, Xingwang Zhao
Abstract: Ensemble clustering has recently emerged a powerful clustering analysis technology for multi-view data. From the existing work, these techniques held great promise, but most of them are inefficient for large data. Some researchers have also proposed efficient ensemble clustering algorithms, but these algorithms devote to data objects with the same feature spaces,rnwhich are not satisfied for multi-view data. To overcome these deficiencies, an efficient ensemble clustering algorithm for multi-view mixed data is developed from the cluster-level perspective. Firstly, a set of clustering solutions are produced with the K-Prototypes clustering algorithm on each view multiple times, respectively. Then, a cluster-cluster similarity matrix is constructed by considering all the clustering solutions. Next, the METIS algorithm is conduct metaclustering based on the similarity matrix. After that, the final clustering results are obtained by applying majority voting to assign the objects to their corresponding clusters based on the metaclustering. The corresponding time complexity of the proposed algorithm is analyzed as well. Experimental results on several multi-view datasets demonstrated the superiority of our proposed algorithm.
Keywords: Multi-view Data; Mixed Data; K-Prototypes Clusteringrn Algorithm; Ensemble Clustering.
Special Issue on: Machine Intelligence for Civil Engineering
Optimal Selecting and Scaling the Earthquake Accelerograms According to Eurocode 8 Using Ranked Particles Optimization
by Amirfarzad Behnam, Amir Nasrollahi
Abstract: In the present study, we aimed at selecting and scaling accelerograms using metaheuristic algorithms. Selecting and scaling accelerograms is an important step in the nonlinear structural analysis. Usually, the accelerograms are selected from the same soil type and they are multiplied by one scaling factor to move its response spectrum above the standard design spectra. In this research, we modified the previous formulation and used ranked particles optimization (RPO) to optimize the process of selecting and scaling. The records were selected from the database including 380 earthquake records from Pacific Earthquake Engineering Research Center (PEER). This database contains most of the strong earthquakes occurred all over the world. For each soil type of Eurocode 8, accelerograms were selected from the same soil type and from the whole database. The results show that regardless of soil types when the accelerograms are selected from the whole database, the cost function has a smaller value. The same methodology can be easily used for other standards by slight modification of the cost and the penalty functions.
Keywords: Accelerograms; Scaling; Ranked Particles Optimization; Eurocode 8; Response Spectra; Design Spectra.