Extending ventilation duration estimations approach from adult toneonatal intensive care patients using artificial neural networks

In earlier work, the research group successfully used artificial neural networks (ANNs) to estimate ventilation duration for adult intensive care unit (ICU) patients. The ANNs performed well in terms of correct classification rate (CCR) and average squared error (ASE) classifying the outcome into two classes: whether patients were ventilated for less than/equal to or for more than 8 h (⩽ or >). The objective of new work was to apply this adult model to the estimation of ventilation with neonatal ICU (NICU) patient records. The performance obtained with the neonatal patients was comparable to that previously found with the adult database, again as measured in terms of a maximum CCR and a minimum ASE. The effectiveness of using the weight-elimination technique in controlling overfitting was again validated for the neonatal patients as it had been for our adult patients. It was concluded that the approach developed for ICU adult patients was also successfully applied to a different medical environment: neonatal ICU patients     

Uncertainty in the output of artificial neural networks

Analysis of the performance of artificial neural networks (ANNs) is usually based on aggregate results on a population of cases. In this paper, we analyze ANN output corresponding to the individual case. We show variability in the outputs of multiple ANNs that are trained and "optimized" from a common set of training cases. We predict this variability from a theoretical standpoint on the basis that multiple ANNs can be optimized to achieve similar overall performance on a population of cases, but produce different outputs for the same individual case because the ANNs use different weights. We use simulations to show that the average standard deviation in the ANN output can be two orders of magnitude higher than the standard deviation in the ANN overall performance measured by the Az value. We further show this variability using an example in mammography where the ANNs are used to classify clustered microcalcifications as malignant or benign based on image features extracted from mammograms. This variability in the ANN output is generally not recognized because a trained individual ANN becomes a deterministic model. Recognition of this variability and the deterministic view of the ANN present a fundamental contradiction. The implication of this variability to the classification task warrants additional study.     

Classification of seismic signals by integrating ensembles ofneural networks

We examine a classification problem in which seismic waveforms of natural earthquakes are to be distinguished from waveforms of man-made explosions. We present an integrated classification machine (ICM), which is a hierarchy of artificial neural networks (ANNs) that are trained to classify the seismic waveforms. In order to maximize the gain of combining the multiple ANNs, we suggest construction of a redundant classification environment (RCE) that consists of several “experts” whose expertise depends on the different input representations to which they are exposed. In the proposed scheme, the experts are ensembles of ANN, trained on different bootstrap replicas. We use various network architectures, different time-frequency decompositions of the seismic waveforms, and various smoothing levels in order to achieve an RCE. A confidence measure for the ensemble's classification is defined based on the agreement (variance) within the ensembles, and an algorithm for a nonlinear integration of the ensembles using this measure is presented. An implementation on a data set of 380 seismic events is described, where the proposed ICM had classified correctly 92% of the testing signals. The comparison we made with classical methods indicates that combining a collection of ensembles of ANNs can be used to handle complex high dimensional classification problems     

A CAD Approach Based on Artificial Neural Networks for Shielded Multilayered Coplanar Waveguides

In this paper a computer-aided design (CAD) approach based on artificial neural networks (ANNs) was successfully introduced to determine the characteristic parameters of shielded multilayered coplanar waveguides (SMCPWs). ANNs are trained with four learning algorithms to obtain better performance and faster convergence with simpler structure. The best results for training and test were obtained from the models trained with Bayesian regularization and Levenberg-Marquardt algorithms. The neural model results are in very good agreement with the results available in the literature for SMCPWs and three other different shielded CPW structures. One can calculate the quasi-static parameters of these four different shielded CPW configurations using only one neural model proposed in this work, easily, simply and accurately.     

Neural models for an asymmetric coplanar stripline with an infinitely wide strip

This paper presents a new approach, based on artificial neural networks (ANNs), to determine the characteristic impedance and the effective permittivity of an asymmetric coplanar stripline (ACPS) with an infinitely wide strip. ANNs are trained with five learning algorithms to obtain better performance and faster convergence with simpler structure. The best results for training and test were obtained from the models trained with the Levenberg–Marquardt and the Bayesian regularization algorithms. The results obtained by using the neural model are in very good agreement with the results available in the literature. The neural models presented in this work provide simplicity and accuracy to determine both the parameters of an ACPS. The method is not time consuming and is easily included in a CAD system.     

ANN Synthesis Models Trained with Modified GA‐LM Algorithm for ACPWs with Conductor Backing and Substrate Overlaying

Accurate synthesis models based on artificial neural networks (ANNs) are proposed to directly obtain the physical dimensions of an asymmetric coplanar waveguide with conductor backing and substrate overlaying (ACPWCBSO). First, the ACPWCBSO is analyzed with the conformal mapping technique (CMT) to obtain the training data. Then, a modified genetic‐algorithm‐Levenberg‐Marquardt (GA‐LM) algorithm is adopted to train ANNs. In the algorithm, the maximal relative error (MRE) is used as the fitness function of the chromosomes to guarantee that the MRE is small, while the mean square error is used as the error function in LM training to ensure that the average relative error is small. The MRE of ANNs trained with the modified GA‐LM algorithm is less than 8.1%, which is smaller than those trained with the existing GA‐LM algorithm and the LM algorithm (greater than 15%). Lastly, the ANN synthesis models are validated by the CMT analysis, electromagnetic simulation, and measurements.     

On the application and design of artificial neural networks formotor fault detection. I

The general design considerations for feedforward artificial neural networks (ANNs) to perform motor fault detection are presented. A few noninvasive fault detection techniques are discussed, including the parameter estimation approach, human expert approach, and ANN approach. A brief overview of feedforward nets and the backpropagation training algorithm, along with its pseudocodes, is given. Some of the neural network design considerations such as network performance, network implementation, size of training data set, assignment of training parameter values, and stopping criteria are discussed. A fuzzy logic approach to configuring the network structure is presented     

Field-oriented control of induction motors using neural-networkdecouplers

This paper presents a novel approach to the field-oriented control (FOC) of induction motor drives. It discusses the introduction of artificial neural networks (ANNs) for decoupling control of induction motors using FOC principles. Two ANNs are presented for direct and indirect FOC applications. The first performs an estimation of the stator flux for direct field orientation, and the second is trained to map the nonlinear behavior of a rotor-flux decoupling controller. A decoupling controller and flux estimator were implemented upon these ANNs using the MATLAB/SIMULINK neural-network toolbox. The data for training are obtained from a computer simulation of the system and experimental measurements. The methodology used to train the networks with the backpropagation learning process is presented. Simulation results reveal some very interesting features and show that the networks have good potential for use as an alternative to the conventional field-oriented decoupling control of induction motors     

Classification of soil texture using remotely sensed brightness temperature over the Southern Great Plains

This study explores the use of artificial neural networks (ANNs) models and brightness temperature from the Southern Great Plains in the United States to classify soil into different textures. Previous studies using ANN models and brightness temperature in a single drying cycle suggested that they might contain sufficient features to classify soil into three categories. To classify soil into more than three groups and to explore the limits of classification accuracy, this paper suggests the use of multiple-drying-cycle brightness temperature data. We have performed several experiments with feedforward neural network (FFNN) models, and the results suggest that the maximum achievable classification accuracy through the use of multiple-drying-cycle brightness temperature is about 80%. It appears that the rapidly changing space-time evolution of brightness temperature will restrict the FFNN model performance. Motivated by these observations, we have used a simple prototype-based classifier, known as the 1-NN model, and achieved 86% classification accuracy for six textural groups. A comparison of error regions predicted by both models suggests that for the given input representation maximum achievable accuracy for classification into six soil texture types is about 93%.     

On the application and design of artificial neural networks formotor fault detection. II

For part I see ibid., vol.40, no.2, p.181-8 (1993). Some neural network design considerations, such as network performance, network implementation, size of training data set, assignment of training parameter values, and stopping criteria, are discussed. A fuzzy logic approach to configuring the network structure is presented, to automate the network design. Successful results are obtained from using artificial neural networks (ANNs) on motor fault detection and fuzzy logic in the network configuration design. It is concluded that these emerging technologies are promising for future widespread industrial usage     

Artificial neural-network-based diagnosis of CVD barrel reactor

This paper presents an artificial neural network (ANN) based diagnostic strategy applied to a chemical vapor deposition (CVD) barrel reactor of the type commonly used in silicon epitaxy. The strategy is based on the spatial variation of the rate of deposition of silicon on a facet of the reactor. Our hypothesis is that this spatial variation, quantified as a vector of variously measured standard deviations, encodes a pattern reflecting the state of the reactor. Therefore, a process fault (event) can be diagnosed by decoding the pattern by an ANN. We implemented this simple scheme by simulating different events by means of a regression model relating the rate of deposition to the process settings. Three different events were simulated and various ANNs were trained to detect and classify these events. It is shown that a single ANN or a combination of ANNs does an excellent job. We also demonstrate that the threshold rule for setting the threshold of a binary output neuron performing a classification task enhances the diagnostic performance. A novel multiple expert scheme that refers to several ANNs trained in the same classification task for decision-making in order to resolve ambiguities and improve the reliability of the final decision is presented and shown to be effective     

Neural network approach to land cover mapping

A pattern classification method is proposed for remote sensing data using neural networks. First, the authors apply the error backpropagation (BP) algorithm to classify the remote sensing data. In this case, the classification performance depends on a training data set. In order to get stable and precise classification results, the training data set is selected based on geographical information and Kohonen's self-organizing feature map. Using the training data set and the error backpropagation algorithm, a layered neural network is trained such that the training patterns are classified with a specified accuracy. After training the neural network, some pixels are deleted from the original training data set if they are incorrectly classified and a new training data set is built up. Once training is complete, a testing data set is classified by using the trained neural network. The classification results of LANDSAT TM data show that this approach produces excellent results which are more realistic and noiseless compared with a conventional Bayesian method     

基于特征聚类的稀疏自编码快速算法

稀疏自编码网络在自然语言、图像处理等领域都取得了显著效果.已有的研究表明增加网络提取的特征个数可以优化稀疏自编码网络的处理效果,同时该操作将导致网络训练耗时过长.为尽可能减少网络的训练时间,本文提出了一种基于特征聚类的稀疏自编码快速算法.本算法首先根据K均值聚类最优数确定本质特征的个数,再由网络训练得到本质特征,并通过旋转扭曲增加特征的多样性,使网络处理效果得到提升的同时,减少网络训练耗间.实验在标准的手写体识别数据库MNIST和人脸数据库CMU-PIE上进行,结果表明本文所提算法能在保证网络正确率有所提升的同时,大幅度缩短网络训练耗时.     

基于递阶粒子群优化径向基函数人工神经网络的光性能监控

为解决差错反向传输神经网络在透明可重构光网络光性能监测中精度不足的问题,提出一种基于优化的径向基函数人工神经网络的光性能监测方案。在该方案中,以信号眼图参数为网络输入,以光信噪比、色散和偏振模色散为网络输出;采用二进制与十进制相结合编码的递阶粒子群方法,用适应度函数引导粒子向小规模和小误差方向运动,进行神经网络的结构与参数自适应优化;分别以不同光信噪比,不同色散和偏振模色散水平仿真信道中传输速率为40 Gb/s差分相移键控仿真信号,进行网络训练和测试,并将测试结果与相同情形下基于差错反向传输法神经网络的光性能监测结果进行比较。结果表明,所提方案在保有人工神经网络方案优点的基础上,有着更好的监测精度。     

Identification of spinal deformity classification with total curvature analysis and artificial neural network

In this paper, a multilayer feed-forward, back-propagation (MLFF/BP) artificial neural network (ANN) was implemented to identify the classification patterns of the scoliosis spinal deformity. At the first step, the simplified 3-D spine model was constructed based on the coronal and sagittal X-ray images. The features of the central axis curve of the spinal deformity patterns in 3-D space were extracted by the total curvature analysis. The discrete form of the total curvature, including the curvature and the torsion of the central axis of the simplified 3-D spine model was derived from the difference quotients. The total curvature values of 17 vertebrae from the first thoracic to the fifth lumbar spine formed a Euclidean space of 17 dimensions. The King classification model was tested on this MLFF/BP ANN identification system. The 17 total curvature values were presented to the input layer of MLFF/BP ANN. In the output layer there were five neurons representing five King classification types. A total of 37 spinal deformity patterns from scoliosis patients were selected. These 37 patterns were divided into two groups. The training group had 25 patterns and testing group had 12 patterns. The 25-pattern training group was further divided into five subsets. Based on the definition of King classification system, each subset contained all five King types. The network training was conducted on these five subsets by the hold-out method, one of cross-validation variants, and the early stop method. In each one of the five cross-validation sessions, four subsets were alternatively used for estimation learning and one subset left was used for validation learning. Final network testing was conducted with remaining 12 patterns in testing group after the MLFF/BP ANN was trained by all five subsets in training group. The performance of the neural network was evaluated by comparing between two network topologies, one with one hidden layer and another with two hidden layers. The results are shown in three tables. The first table shows network errors in estimation learning and the second table shows identification rates in validation learning. The network errors and identification rates in the last round of network training and testing are shown in the third table. Each table has a comparison for both one hidden layer and two hidden layer networks.     

Co-Occurrence Histogram Based Ensemble of Classifiers for Classification of Cervical Cancer Cells

To explore the potential of conventional image processing techniques in the classification of cervical cancer cells, in this work, a co-occurrence histogram method was employed for image feature extraction and an ensemble classifier was developed by combining the base classifiers, namely, the artificial neural network (ANN), random forest (RF), and support vector machine (SVM), for image classification. The segmented pap-smear cell image dataset was constructed by the k-means clustering technique and used to evaluate the performance of the ensemble classifier which was formed by the combination of above considered base classifiers. The result was also compared with that achieved by the individual base classifiers as well as that trained with color, texture, and shape features. The maximum average classification accuracy of 93.44% was obtained when the ensemble classifier was applied and trained with co-occurrence histogram features, which indicates that the ensemble classifier trained with co-occurrence histogram features is more suitable and advantageous for the classification of cervical cancer cells.     

Positioning in WLAN environment by use of artificial neural networks and space partitioning

Short range wireless technologies such as wireless local area network (WLAN), Bluetooth, radio frequency identification, ultrasound and Infrared Data Association can be used to supply position information in indoor environments where their infrastructure is deployed. Due to the ubiquitous presence of WLAN networks, positioning techniques in these environments are the scope of intense research. In this paper, the position determination by the use of artificial neural networks (ANNs) is explored. The single ANN multilayer feedforward structure and a novel positioning technique based on cascade-connected ANNs and space partitioning are presented. The proposed techniques are thoroughly investigated on a real WLAN network. Also, an in-depth comparison with other well-known techniques is shown. Positioning with a single ANN has shown good results. Moreover, when utilising space partitioning with the cascade-connected ANNs, the median error is further reduced for as much as 28%.     

Medical diagnosis with C4.5 rule preceded by artificial neural network ensemble

Comprehensibility is very important when machine learning techniques are used in computer-aided medical diagnosis. Since an artificial neural network ensemble is composed of multiple artificial neural networks, its comprehensibility is worse than that of a single artificial neural network. In this paper, C4.5 Rule-PANE, which combines an artificial neural network ensemble with rule induction by regarding the former as a preprocess of the latter, is proposed. At first, an artificial neural network ensemble is trained. Then, a new training data set is generated by feeding the feature vectors of original training instances to the trained ensemble and replacing the expected class labels of original training instances with the class labels output from the ensemble. Additional training data may also be appended by randomly generating feature vectors and combining them with their corresponding class labels output from the ensemble. Finally, a specific rule induction approach, i.e., C4.5 Rule, is used to learn rules from the new training data set. Case studies on diabetes, hepatitis , and breast cancer show that C4.5 Rule-PANE could generate rules with strong generalization ability, which benefits from an artificial neural network ensemble, and strong comprehensibility, which benefits from rule induction.     

Macrocell path loss prediction using artificial intelligence techniques

The prediction of propagation loss is a practical non-linear function approximation problem which linear regression or auto-regression models are limited in their ability to handle. However, some computational Intelligence techniques such as artificial neural networks (ANNs) and adaptive neuro-fuzzy inference systems (ANFISs) have been shown to have great ability to handle non-linear function approximation and prediction problems. In this study, the multiple layer perceptron neural network (MLP-NN), radial basis function neural network (RBF-NN) and an ANFIS network were trained using actual signal strength measurement taken at certain suburban areas of Bauchi metropolis, Nigeria. The trained networks were then used to predict propagation losses at the stated areas under differing conditions. The predictions were compared with the prediction accuracy of the popular Hata model. It was observed that ANFIS model gave a better fit in all cases having higher R² values in each case and on average is more robust than MLP and RBF models as it generalises better to a different data.     

High-quality voice conversion system based on GMM statistical parameters and RBF neural network

A voice conversion （VC） system was designed based on Gaussian mixture model （GMM） and radial basis function （RBF） neural network. As a voice conversion model, RBF network needs quantities of training data to improve its performance. For one speech, the networks trained by different segments of data have different transformation effects. Since trying segment by segment to obtain the best conversion effect is complex, a conversion method was proposed, that uses GMM for statistics before training RBF network to aim at the problem. The speech transformation and representation using adaptive interpolation of weighted spectrum （STRAIGHT） model is used for accurate extraction of vocal tract spectrum. Then GMM is used to classify the numerous spectral parameters. The obtained mean parameters were trained in RBF network. Experiment reveals that, the soft classification ability of GMM can promptly realize the reduction and classification of training data under the premise of ensuring the training effect. The selection complexity is decreased thereafter. Compared to the conventional RBF network training methods, this method can make the transformation of spectral parameters more effective and improve the quality of converted speech.