Complex-valued Self-regulatory Resource Allocation Network (CSRAN)

In situations with missing data, statistical analyses are usually limited to subjects with complete data. However, such estimates may be biased. The method of 'filling in' missing data is called imputation. This article aimed to present a multiple imputation method. From a data set of 470 surgical patients, logistic models were developed for death as the outcome. Two incomplete data sets were generated: one with 5% and another with 20% of missing data in a single variable. Logistic models were fitted for the complete and incomplete data sets and for the data set completed by multiple imputations. Estimates obtained for the data set with missing data were different from those observed in the complete data set, mainly in the situation with 20% of missing data. The multiple imputation used here appeared efficient, producing very similar results to those obtained with the complete data set. However, one coefficient became non-significant. The analysis using multiple imputations was considered superior to using the data sets that excluded incomplete cases from the analysis.

Exploiting the information content offered by geoelectric data in an efficient manner requires careful selection of the electrode configurations to be used. This can be achieved using sequential experimental design techniques proposed over the past few years. However, these techniques become impractical when large-scale 2D or 3D experiments have to be designed. Even if sequential experimental design were applicable, acquisition of the resulting data sets would require an unreasonably large effort using traditional multielectrode arrays. We present a new, fully parallelized pole-bipole measuring strategy by which large amounts of data can be acquired swiftly. Furthermore, we introduce a new experimental design concept that is based on “complete” data sets in terms of linear independence. Complete data sets include a relatively small number of basis electrode configurations, from which any other configuration can be reconstructedby superposition. The totality of possible configurations is referred to as the comprehensive data set. We demonstrate the benefits of such reconstructions using eigenvalue analyses for the case of noise-free data. In the presence of realistic noise, such reconstructions lead to unstable results when only four-point (bipole-bipole) configurations are considered. In contrast, complete three-point (pole-bipole) data sets allow more stable reconstructions. Moreover, complete pole-bipole data sets can be acquired very efficiently with a fully parallelized system. Resolution properties of complete pole-bipole data sets are illustrated using both noise-free and noisy synthetic data sets. We also show results from a field survey performed over a buried waste disposal site, which further demonstrates the usefulness of our approach. Although this paper is restricted to 2D examples, it is trivial to extend the concept to 3D surveys, where the advantages of parallelized pole-bipole data acquisition become very significant.

Abstract: Machine learning can extract desired knowledge from training examples and ease the development bottleneck in building expert systems. Most learning approaches derive rules from complete and incomplete data sets. If attribute values are known as possibility distributions on the domain of the attributes, the system is called an incomplete fuzzy information system. Learning from incomplete fuzzy data sets is usually more difficult than learning from complete data sets and incomplete data sets. In this paper, we deal with the problem of producing a set of certain and possible rules from incomplete fuzzy data sets based on rough sets. The notions of lower and upper generalized fuzzy rough approximations are introduced. By using the fuzzy rough upper approximation operator, we transform each fuzzy subset of the domain of every attribute in an incomplete fuzzy information system into a fuzzy subset of the universe, from which fuzzy similarity neighbourhoods of objects in the system are derived. The fuzzy lower and upper approximations for any subset of the universe are then calculated and the knowledge hidden in the information system is unravelled and expressed in the form of decision rules.

In recent years, machine learning and knowledge discovery techniques have attracted a great deal of attention in the information area. Classification is one of the important research topics on these research areas. Most of the researches on classification concern that a complete data set is given as a training set and the test data know all values of attributes clearly. Unfortunately, incomplete data are commonly seen in real-world applications. In this paper, we propose a new strategy to deal with the incomplete quantitative data and introduce a supervised learning method based on genetic programming to handle the classification problem with incomplete data in the attributes. Two experiments are designed to evaluate the effectiveness of the proposed approaches.

The combination of music teaching and computer music technology in Colleges and universities and the practical application and operation reform of music theory is a qualitative change improvement to the shortcomings of the traditional music teaching mode. At the same time, it can effectively stimulate college students’ interest in learning music and improve the teaching quality. In the recommendation system, the recommendation algorithm based on matrix decomposition is one of the hot research topics. However, the recommendation accuracy of the common matrix decomposition algorithm is low. In order to improve this problem, this paper takes the potential factor model (LFM) optimization of matrix decomposition algorithm as the research object, and analyzes the shortcomings of the two basic recommendation algorithms in LFM in the optimization speed and recommendation accuracy, Then two improved algorithms are proposed: batch learning algorithm with impulse and hybrid learning algorithm. Finally, through the experimental data test, the recommendation effect of different algorithms is compared. The results show that the performance of the improved algorithm is better. This paper explores the application of multimedia computer music teaching system and harmony teaching course in Colleges and universities, and proves the superiority of computer music technology in the reform of college music teaching mode.KeywordsComputer music teaching systemMatrix decompositionLatent factor modelRecommendation algorithm

Electricity theft-induced power loss is a pressing issue in both traditional and smart grid environments. In smart grids, smart meters can be used to track power consumption behaviour and detect any suspicious activity. However, smart meter readings can be compromised by deploying intrusion tactics or launching cyber attacks. In this regard, machine learning models can be used to assess the daily consumption patterns of customers and detect potential electricity theft incidents. Whilst existing research efforts have extensively focused on batch learning algorithms, this paper investigates the use of online machine learning algorithms for electricity theft detection in smart grid environments, based on a recently proposed dataset. Several algorithms including Naive Bayes, K-nearest Neighbours, K-nearest Neighbours with self-adjusting memory, Hoeffding Tree, Extremely Fast Decision Tree, Adaptive Random Forest and Leveraging Bagging are considered. These algorithms are evaluated using an online machine learning platform considering both binary and multi-class theft detection scenarios. Evaluation metrics include prediction accuracy, precision, recall, F-1 score and kappa statistic. Evaluation results demonstrate the ability of the Leveraging Bagging algorithm with an Adaptive Random Forest base classifier to surpass all other algorithms in terms of all the considered metrics, for both binary and multi-class theft detection. Hence, it can be considered as a viable option for electricity theft detection in smart grid environments.

SummaryAccurate prediction of the solubility of chemical substances in solvents remains a challenge. The sparsity of high-quality solubility data is recognized as the biggest hurdle in the development of robust data-driven methods for practical use. Nonetheless, the effects of the quality and quantity of data on aqueous solubility predictions have not yet been scrutinized. In this study, the roles of the size and the quality of data sets on the performances of the solubility prediction models are unraveled, and the concepts of actual and observed performances are introduced. In an effort to curtail the gap between actual and observed performances, a quality-oriented data selection method, which evaluates the quality of data and extracts the most accurate part of it through statistical validation, is designed. Applying this method on the largest publicly available solubility database and using a consensus machine learning approach, a top-performing solubility prediction model is achieved.

Semi-supervised learning is especially important in data mining applications because it can make use of plentiful unlabeled data to train the high-quality learning models. Semi-Supervised Support Vector Machine (S3VM) is a powerful semi-supervised learning model. However, the high computational cost and non-convexity severely impede the S3VM method in large-scale applications. Although several learning algorithms were proposed for S3VM, scaling up S3VM is still an open problem. To address this challenging problem, in this paper, we propose a new incremental learning algorithm to scale up S3VM (IL-S3VM) based on the path following technique in the framework of Difference of Convex (DC) programming. The traditional DC programming based algorithms need multiple outer loops and are not suitable for incremental learning, and traditional path following algorithms are limited to convex problems. Our new IL-S3VM algorithm based on the path-following technique can directly update the solution of S3VM to converge to a local minimum within one outer loop so that the efficient incremental learning can be achieved. More importantly, we provide the finite convergence analysis for our new algorithm. To the best of our knowledge, our new IL-S3VM algorithm is the first efficient path following algorithm for a non-convex problem (i.e., S3VM) with local minimum convergence guarantee. Experimental results on a variety of benchmark datasets not only confirm the finite convergence of IL-S3VM, but also show a huge reduction of computational time compared with existing batch and incremental learning algorithms, while retaining the similar generalization performance.

Rolling element bearings constitute the key parts on rotating machinery and their fault diagnosis are of great importance. In this paper, a new intelligent fault diagnosis scheme based on Wavelet Packet Transform (WPT) and Extreme Learning Machine (ELM) is proposed. 16-dimensional wavelet packet node energies were extracted from the original datasets as the feature vector input to the classifiers. A novel classifier, ELM, and its variants, error-minimized ELM (EM-ELM) and online sequential ELM (OS-ELM), were introduced in this study to diagnose the fault on bearings. ELM has been proved to be extremely fast and can provide good generalization performance on many pattern recognition cases. However, preliminary ELM is a batch learning algorithm with a fixed network structure. EM-ELM and OS-ELM are the extends of the preliminary ELM to allow network to grow in the learning process and to learn data sequentially. ELM, EM-ELM and OS-ELM classifiers were evaluated on 13 fault datasets and the empirical results showed that they are really fast and perform well on all the 13 datasets.

Incomplete data with blockwise missing patterns are commonly encountered in analytics, and solutions typically entail listwise deletion or imputation. However, as the proportion of missing values in input features increases, listwise or columnwise deletion leads to information loss, whereas imputation diminishes the integrity of the training data set. We present the blockwise reduced modeling (BRM) method for analyzing blockwise missing patterns, which adapts and improves on the notion of reduced modeling proposed by Friedman, Kohavi, and Yun in 1996 as lazy decision trees. In contrast to the original idea of reduced modeling of delaying model induction until a prediction is required, our method is significantly faster because it exploits the blockwise missing patterns to pretrain ensemble models that require minimum imputation of data. Models are pretrained over the overlapping subsets of an incomplete data set that contain only populated values. During prediction, each test instance is mapped to one of these models based on its feature-missing pattern. BRM can be applied to any supervised learning model for tabular data. We benchmark the predictive performance of BRM using simulations of blockwise missing patterns on three complete data sets from public repositories. Thereafter, we evaluate its utility on three data sets with actual blockwise missing patterns. We demonstrate that BRM is superior to most existing benchmarks in terms of predictive performance for linear and nonlinear models. It also scales well and is more reliable than existing benchmarks for making predictions with blockwise missing pattern data. History: Maytal Saar-Tsechansky served as the senior editor for this article. Data Ethics & Reproducibility Note: The code capsule is available on Code Ocean at https://codeocean.com/capsule/0274716/tree and in the e-Companion to this article (available at https://doi.org/10.1287/ijds.2022.9016 ).

Complementarity-based reserve selection algorithms efficiently prioritize sites for biodiversity conservation, but they are data-intensive and most regions lack accurate distribution maps for the majority of species. We explored implications of basing conservation planning decisions on incomplete and biased data using occurrence records of the plant family Proteaceae in South Africa. Treating this high-quality database as 'complete', we introduced three realistic sampling biases characteristic of biodiversity databases: a detectability sampling bias and two forms of roads sampling bias. We then compared reserve networks constructed using complete, biased, and randomly sampled data. All forms of biased sampling performed worse than both the complete data set and equal-effort random sampling. Biased sampling failed to detect a median of 1-5% of species, and resulted in reserve networks that were 9-17% larger than those designed with complete data. Spatial congruence and the correlation of irreplaceability scores between reserve networks selected with biased and complete data were low. Thus, reserve networks based on biased data require more area to protect fewer species and identify different locations than those selected with randomly sampled or complete data.

Objective: Both the central nervous system and the autonomic nervous system are complex physiological networks which modulate the heart rate. They are spatially extended, have built-in delays and work on many time scales simultaneously—nonhomogeneous networks with multifractal dynamics. The object of our research was the analysis of human heart rate variability (HRV) using the nonlinear multiscale multifractal analysis (MMA) method for several cardiovascular diseases. The analysis of HRV (night-time recordings) involved six groups of patients: 61 healthy persons, 104 cases with aortic valve stenosis, 42 with hypertrophic cardiomyopathy, 36 with atrial fibrillation, 70 patients with coronary artery disease and 19 with congestive heart failure. 85% of patients formed a training data set (282 subjects) and 15% formed a test data set (50 subjects). Approach: Multiscale multifractal analysis allows one to analyze the complexity of HRV and find the scaling properties of its fluctuations. The main result of MMA is the Hurst surface, the shape of which changes depending on the medical case analyzed. We prepared six criteria to distinguish a multifractal pattern for healthy subjects. We also prepared additional criteria, enabling one to recognize atrial fibrillation. Main results: For the training data set, we obtained the following accuracy statistics in distinguishing the patients from the healthy: 68% for coronary artery disease, 67% for hypertrophic cardiomyopathy, 88% for atrial fibrillation, 74% for aortic valve stenosis and 83% for congestive heart failure. For the complete training data set we obtained an accuracy of 73%, and 80% for the test data set (mean for ten random selections of the test data set). Significance: The results of MMA presented here provide an additional input into the diagnostic process and may help to create a paradigm for future studies on medical screening methods, especially in that MMA focuses on very low frequencies of HRV not easily accessible by standard medical techniques. Satisfactory statistics for screening using both MMA and the unfiltered version of LF/HF indicate that the nature of the complete network moderating heart rhythm needs to be studied and that sinus rhythm in clinical patients may not always be separated from arrhythmia when its incidence is large.

This paper introduces several nonlinear multi-model ensemble techniques for multiple chaotic models in high-dimensional phase space by means of artificial neural networks. A chaotic model is built by way of the time-delayed phase space reconstruction of the time series from observables. Several predictive global and local models, including Multi-layered Perceptron Neural Network (MLP-NN), are constructed and a number of multi-model ensemble techniques are implemented to produce more accurate hybrid models. One of these techniques is the nonlinear multi-model ensemble using one kind of dynamic neural network so-called Focused Time Delay Neural Network (FTDNN) with batch and incremental learning algorithms. The proposed techniques were used and tested for predicting storm surge dynamics in the North Sea. The results showed that the accuracy of multi-model ensemble predictions is generally improved in comparison to the one by single models. An FTDNN with incremental learning is more desirable for real-time operation, however in our experiments it was less accurate than batch learning.

Background/Objectives: This paper deals with the handling method of incomplete training data that is partially lost. It is unavoidable in a ubiquitous environment that deals with data collected from multiple devices or from a long distance, but it is difficult to expect good results due to loss of information if the lost part is discarded. Methods/Statistical analysis: Various algorithms have been proposed to solve this problem. Among them, the algorithm for learning by changing the format of the training data to fit incomplete data has been applied to various problems and has shown good results. This data format conversion method is called a data expansion technique, and has two characteristics: the importance can be adjusted differently for each event, and a probability value can be assigned to each cardinality of each variable. Findings: The second feature, the ability to assign a probability value, was used to assign a compensation value for loss data. The first attempt to do this was to assign equal probability values to the loss values. In the classification algorithm using the entropy function, the variable containing more loss values is not selected in the upper node. However, the method of allocating the equal value had the point that the original information is ignored. Therefore, a method of obtaining the entropy probability with complete information excluding the loss value and filling it in the loss value was also proposed. This paper starts with the basic idea that this method is further developed and the lost information can be found in the area classified by the classification algorithm. In terms of implementation, the training data is divided into two parts: lost and non-lost events, and then classified using the C4.5 classification algorithm with the data that is not lost, and each classification area is obtained. Then, by using the information remaining in the event of the loss data, the classification area is sequentially traversed according to a predetermined algorithm to find the area closest to the loss event. And after expressing the value of this area as a probability, the loss value is replaced with this probability value, which means compensation for the loss value. Improvements/Applications: After compensating for the total loss values, the recovered training data is learned with one of the SVM algorithms in order to evaluate the performance of how much information leakage has been preserved, and its performance is compared. As a result of the experiment with different degrees of loss for each variable, it was confirmed that the loss of information can be minimized and used as a method of meaningful compensation.

Hydroxynitrile lyase from M. esculenta (cassava) was crystallized in two different crystal forms by the hanging-drop vapour-diffusion method. Crystals of form I were obtained from a mixture of polyethylene glycol 8000 and 2-methyl-2,4-pentanediol, and belong to the tetragonal space group P41212 or its enantiomorph P43212, with unit-cell parameters a = b = 105.9, c = 188.9 A and with two molecules in the asymmetric unit. These crystals diffract to 2.9 A with conventional X-ray sources and beyond 2.1 A resolution with synchrotron radiation. The crystals are relatively sensitive to radiation damage and conditions for flash-cooling the crystals have been established. A complete native data set has been collected up to 2.2 A resolution. Crystal form II has been obtained at pH 5.6 using lithium sulfate as a precipitant. The crystals apparently belong to the orthorhombic space group P21212, with unit-cell parameters a = 117.52, b = 127.09 and c = 78.08 A, have two molecules in the asymmetric unit and diffract to beyond 2.0 A resolution. A complete native data set has been collected to 2.2 A resolution.