Adverse effects of tribromoethanol as used in the production of transgenic mice

The authors developed and cross-validated prediction models for newly diagnosed cases of liver disorders by using logistic regression and neural networks. Computerized files of health care encounters from the Fallon Community Health Plan were used to identify 1,674 subjects who had had liver-related health services between July 1, 1992, and June 30, 1993. A total of 219 subjects were confirmed by review of medical records as incident cases. The 1,674 subjects were randomly and evenly divided into training and test sets. The training set was used to derive prediction algorithms based solely on the automated data; the test set was used for cross-validation. The area under the Receiver Operating Characteristic curve for a neural network model was significantly larger than that for logistic regression in the training set (p = 0.04). However, the performance was statistically equivalent in the test set (p = 0.45). Despite its superior performance in the training set, the generalizability of the neural network model is limited. Logistic regression may therefore be preferred over neural network on the basis of its established advantages. More generalizable modeling techniques for neural networks may be necessary before they are practical for medical research.     

基于Kmeans–BP神经网络的KR工序终点铁水硫含量预测模型

针对KR工序终点铁水硫含量预测问题,提出一种基于Kmeans聚类分析和BP神经网络（BPNN）相结合的建模方法。首先,通过Kmeans聚类对KR工序生产数据进行模式识别和分类,构建不同工况特征的数据集;然后,基于BP神经网络,针对不同数据集训练预测模型;最后,将不同数据集的预测模型进行集成,形成最终的终点铁水硫含量预测模型,实现对不同铁水条件和工况条件的预测。利用某钢铁企业实际生产数据,分别用基于脱硫反应动力学、BP神经网络和Kmeans–BPNN方法建立的预测模型,对KR工序终点铁水硫含量进行预测。结果表明,Kmeans–BPNN的KR工序终点硫含量预测模型的精度显著高于脱硫反应动力学和BP神经网络的预测模型。      

Effectiveness of Different Artificial Neural Network Training Algorithms in Predicting Protozoa Risks in Surface Waters

A neural network approach was employed to relate risky Cryptosporidium and Giardia concentrations with other biological, chemical and physical parameters in surface water. A set of drinking water samples was classified as “risky” and “nonrisky” based on the concentrations of full and empty oocysts, and cycsts of Cryptosporidium and Giardia, respectively. Given the constraints associated with collecting large sets of microbial data, the study was aimed at identifying an effective training algorithm that would maximize the performance of a neural network model working with a relatively small dataset. A number of algorithms for training neural networks, including gradient search with first- and second-order partial derivatives, and genetic search were used and compared. Results showed that genetic algorithm based neural network training consistently provided better results compared to other training methods.     

Modeling and Predicting Biological Performance of Contact Stabilization Process Using Artificial Neural Networks

In this paper, the microfauna distribution data of a contact stabilization process were used in a neural network system to model and predict the biological activity of the effluent. Five uncorrelated components of the microfauna were used as the artificial neural network model input to predict the dehydrogenase activity of the effluent (DAE) using back-propagation and general regression algorithms. The models’ optimum architectures were determined for the back-propagation neural network (BPNN) model by varying the number of hidden layers, hidden transfer functions, test set size percentages, and initial weights. Comparison of the two model prediction results showed that the genetic general regression neural network model demonstrated the ability to calibrate the multicomponent microfauna, and yielded reliable DAE close to that resulting from direct experimentation, and thus was judged superior to BPNN models.     

Recurrent respiratory papillomatosis

PURPOSE: To optimize the performance of artificial neural networks in the prediction of pulmonary embolism from ventilation-perfusion (V-P) scans. MATERIALS AND METHODS: Neural networks were constructed with a set of V-P scan criteria that included sharpness and completeness of perfusion defects and involved quantification of abnormalities by using a continuous numeric scale. Several network parameters were systematically varied. Networks were trained with 150 cases and tested with 30 different cases. Findings were compared with those of pulmonary angiography. RESULTS: Networks capable of performing as well as experienced nuclear medicine physicians could be constructed with few V-P scan features. A brief training period was optimal (50-100 iterations). Further training diminished network performance. CONCLUSION: Effective neural networks can be constructed by using a limited number of unconventional V-P scan features. Several parameters can be adjusted to optimize performance.     

Estimation of Aeroelastic Parameters of Bridge Decks Using Neural Networks

A new method of estimating flutter derivatives using artificial neural networks is proposed. Unlike other computational fluid dynamics based numerical analyses, the proposed method estimates flutter derivatives utilizing previously measured experimental data. One of the advantages of the neural networks approach is that they can approximate a function of many dimensions. An efficient method has been developed to quantify the geometry of deck sections for neural network input. The output of the neural network is flutter derivatives. The flutter derivatives estimation network, which has been trained by the proposed methodology, is tested both for training sets and novel testing sets. The network shows reasonable performance for the novel sets, as well as outstanding performance for the training sets. Two variations of the proposed network are also presented, along with their estimation capability. The paper shows the potential of applying neural networks to wind force approximations.     

Application of multilayer feed-forward neural networks to automated compound identification in low-resolution open-path FT-IR spectrometry

A drawback of current open-path Fourier transform infrared (OP/FT-IR) systems is that they need a human expert to determine those compounds that may be quantified from a given spectrum. In this work, multilayer feed-forward neural networks with one hidden layer were used to automatically recognize compounds in an OP/FT-IR spectrum without compensation of absorption lines due to atmospheric H2O and CO2. The networks were trained by fast-back-propagation. The training set comprised spectra that were synthesized by digitally adding randomly scaled reference spectra to actual open-path background spectra measured over a variety of path lengths and temperatures. The reference spectra of 109 compounds were used to synthesize the training spectra. Each neural network was trained to recognize only one compound in the presence of up to 10 other interferences in an OP/FT-IR spectrum. Every compound in a database of vaporphase reference spectra can be encoded in an independent neural network so that a neural network library can be established. When these networks are used for the identification of compounds, the process is analogous to spectral library searching. The effect of learning rate and band intensities on the convergence of network training was examined. The networks were successfully used to recognize five alcohols and two chlorinated compounds in field-measured controlled-release OP/FT-IR spectra of mixtures of these compounds.     

Robust Training Termination Criterion for Back-Propagation ANNs Applicable to Small Data Sets

One of the daunting tasks of a neural network modeler is prescribing an appropriate training termination criterion, a criterion that avoids underfitting or overfitting the underlying functional relationship between input and output variables. This is particularly true when dealing with smaller data sets that do not offer the luxury of splitting the database into traditional training, testing, and validation sets. In the absence of a testing data set or when the testing data set is small, which is not very uncommon when working with environmental databases, it is extremely difficult to know when to terminate the training exercise. This paper proposes a new criterion that provides adequate guidance on training termination without the necessity for a testing data set and illustrates the validity of the proposed criterion on three data sets for water resources and environmental engineering applications. An extensive study of a number of large and small data sets has indicated that the moving average of relative strength index of a randomly generated dummy input variable tends to reach zero at the optimal termination point and tends to move away from zero beyond the optimal point. Based on this observation, a training terminating index was developed, tested, and validated on three datasets.     

基于SE注意力机制的废钢分类评级方法

为了解决传统人工方法对废钢分类评级人为因素干扰大且效率低下等问题,提出基于挤压?激励（Squeeze?Excitation,SE）注意力机制构建废钢分类评级的深度学习网络模型,并对采集到的废钢卸载过程图像进行模型训练和验证。首先,搭建物理尺寸比例为1∶3废钢质量查验物理模型,采用高分辨率视觉传感器模拟采集货车卸载废钢作业场景下不同废钢的形貌特征;然后,对采集到的废钢图像使用跨阶段局部网络进行特征提取,利用空间金字塔结构解决特征丢失问题,采用注意力机制关注通道间的相关性;最后,在包含7个标签分类的两个数据集进行模型训练与验证。实验表明:该模型能够有效地对不同级别的废钢进行自动评级判定,全类别准确率达到83.7%,全类别平均精度为88.8%,在准确性方面相比于传统人工验质方法具有显著优势,解决了废钢入库过程中质量评价的公正性难题。      

Fuzzy C-means Rule Generation for Fuzzy Entry Temperature Prediction in a Hot Strip Mill

Variable estimation for finishing mill set-up in hot rolling is greatly affected by measurement uncertainties, variations in the incoming bar conditions and product changes.The fuzzy C-means algorithm was evaluated for rule-base generation for fuzzy and fuzzy grey-box temperature estimation.Experimental data were collected from a real-life mill and three different sets were randomly drawn.The first set was used for rule-generation,the second set was used for training those systems with learning capabilities,while the third one was used for validation.The perform-ance of the developed systems was evaluated by five performance measures applied over the prediction error with the validation set and was compared with that of the empirical rule-base fuzzy systems and the physical model used in plant.The results show that the fuzzy C-means generated rule-bases improve temperature estimation;however,the best results are obtained when fuzzy C-means algorithm,grey-box modeling and learning functions are combined. Application of fuzzy C-means rule generation brings improvement on performance of up to 72%.     

Neural Network Modeling of Resilient Modulus Using Routine Subgrade Soil Properties

Artificial neural network (ANN) models are developed in this study to correlate resilient modulus with routine properties of subgrade soils and state of stress for pavement design application. A database is developed containing grain size distribution, Atterberg limits, standard Proctor, unconfined compression, and resilient modulus results for 97 soils from 16 different counties in Oklahoma. Of these, 63 soils (development data set) are used in training, and the remaining 34 soils (evaluation data set) from two different counties are used in the evaluation of the developed models. A commercial software, STATISTICA 7.1, is used to develop four different feedforward-type ANN models: linear network, general regression neural network, radial basis function network, and multilayer perceptrons network (MLPN). In each of these models, the input layer consists of seven nodes, one node for each of the independent variables, namely moisture content (w), dry density (γd), plasticity index (PI), percent passing sieve No. 200 (P200), unconfined compressive strength (Uc), deviatoric stress (σd), and bulk stress (θ). The output layer consists of only one node—resilient modulus (MR). After the architecture is set, the development data set is fed into the model for training. The strengths and weaknesses of the developed models are examined by comparing the predicted MR values with the experimental values with respect to the R2 values. Overall, the MLPN model with two hidden layers was found to be the best model for the present development and evaluation data sets. This model as well as the other models could be refined using an enriched database.     

基于神经网络的真空感应炉终点钢水温度预报

针对真空感应炉生产过程中温度测量成本较高及精度较差等不足,建立了基于RBF神经网络的真空感应炉终点钢水温度预报模型。对输入参数作了详细的分析、筛选,并运用聚类算法对该模型进行了训练。结合现场数据进行了学习和预报,预报命中率较高,表明采用该方法可很好地预报钢水温度。     

Sensitivity Analysis of Neural Networks in Spool Fabrication Productivity Studies

The back-propagation neural network (BPNN) has been researched and applied as a convenient decision-support tool in a variety of application areas in civil engineering. However, learning algorithms such as the BPNN do not give information on the effect of each input parameter or influencing variable upon the predicted output variable. The model's sensitivity to changes in its parameters is generally probed by testing the response of a mature network on various input scenarios. In this paper, the relationships between an output variable and an input parameter are sorted out based on the BPNN algorithm. The input sensitivity of the BPNN is defined in exact mathematical terms in light of both normalized and raw data. The difference between a BPNN and regression analysis of statistics is discussed, and the sophistication and superiority of the BPNN over regression analysis is further demonstrated in a case study based on a small data set. In addition, statistical analysis of input sensitivity based on Monte Carlo simulation enables the modeler to understand the rationale of a BPNN's reasoning and have preknowledge about the effectiveness of model implementation in a probabilistic fashion. The sensitivity analysis of the BPNN is successfully applied to analyze the labor production rate of pipe spool fabrication in a real industrial setting. Important aspects of the application, including problem definition, factor identification, data collection, and model testing based on real data, are discussed and presented.     

Neural network classification of mutagens using structural fragment data

A neural network was applied to a large, structurally heterogeneous data set of mutagens and non-mutagens to investigate structure-property relationships. Substructural data comprising a total of 1280 fragments were used as inputs. The training of the back-propagation networks was directed by an algorithm which selected an optimal subset of fragments in order to maximize their discriminating power, and a good predictive network. The system comprised three programs: the first used a keyfile of 100 fragments to generate training and test files, the second was the network itself and a procedure for ranking the effectiveness of these fragments and the third randomly replaced the lowest fragments. This cycle was then repeated. After running on a 386/33 PC several networks produced approximately 11% failures in the test set and 6% in the training set. By simplifying the output of the hidden layer it was possible to describe the hidden layer states in terms of clusters of mutagens and non-mutagens. Some of these clusters were structurally homogeneous and contained known mutagenic and non-mutagenic structural classes. This analysis provided a useful means of demonstrating how the network was classifying the data.     

基于神经网络与LabVIEW平台建立焊锡真空炉粗锡含铅量的软测量模型与在线监测研究

焊锡真空炉粗锡含Pb量的高低直接关系到焊锡真空炉的生产效率,为了改变目前粗锡含Pb量只能通过人工化验才能得到的现状,实验基于反向传播神经网络(Back-Propagation Neural Network,BPNN)与广义回归神经网络(Generalized Regression Neural Network,GRNN)算法原理,构建了BPNN与GRNN软测量模型并对这两种模型的预测效果进行了对比分析,结果表明基于GRNN的粗锡含Pb量软测量模型具有较高的预测精度。同时,采用虚拟仪器(LabVIEW)中的Matlab Script节点技术,成功开发了基于LabVIEW的粗锡含Pb量监测系统,实现了基于BPNN与GRNN软测量模型的粗锡含Pb量实时在线软预测,运行结果表明所开发的监测系统运行稳定可靠。     

Aircraft Icing Detection, Identification, and Reconfigurable Control Based on Kalman Filtering and Neural Networks

Flight in all weather conditions has necessitated correctly detecting icing and taking reasonable measures against it. This work aims at the detection and identification of airframe icing based on statistical properties of aircraft dynamics and reconfigurable control protecting aircraft from hazardous icing conditions. A Kalman filter is used for the data collection for the detection of icing, which aerodynamically deteriorates flight performance. A neural network process is applied for the identification of icing model of the aircraft, which is represented by five parameters based on past experiments for iced wing airfoils. Icing is detected by a Kalman filtering innovation sequence approach. A neural network structure is embodied such that its inputs are the aircraft estimated measurements and its outputs are the parameters affected by ice, which corresponds to the aircraft inverse dynamic model. The necessary training and validation set for the neural network model of the iced aircraft are obtained from the simulations of nominal model, which are performed for various icing conditions. In order to decrease noise effects on the states and to increase training performance of the neural network, the estimated states by the Kalman filter are used. A suitable neural network model of aircraft inverse dynamics is obtained by using system identification methods and learning algorithms. This trained model is used as an application for the control of the aircraft that has lost its controllability due to icing. The method is applied to F16 military and A340 commercial aircraft models and the results seem to be good enough.     

Relationship of actin, microtubules, and crosswall synthesis during septation in Aspergillus nidulans

A significant limitation of neural networks is that the representations they learn are usually incomprehensible to humans. We have developed an algorithm, called TREPAN, for extracting comprehensible, symbolic representations from trained neural networks. Given a trained network, TREPAN produces a decision tree that approximates the concept represented by the network. In this article, we discuss the application of TREPAN to a neural network trained on a noisy time series task: predicting the Dollar-Mark exchange rate. We present experiments that show that TREPAN is able to extract a decision tree from this network that equals the network in terms of predictive accuracy, yet provides a comprehensible concept representation. Moreover, our experiments indicate that decision trees induced directly from the training data using conventional algorithms do not match the accuracy nor the comprehensibility of the tree extracted by TREPAN.     

Steel annealing furnace robust neural network model

Abstract

In this day and age, galvanised coated steel is an essential product in several key manufacturing sectors because of its anticorrosive properties. The increase in demand has led managers to improve the different phases in their production chains. Among the efforts needed to accomplish this task, process modelling can be identified as the one with the most powerful outputs in spite of its non-trivial development. In many fields, such as industrial modelling, multilayer feedforward neural networks are often proposed as universal function approximators. These supervised neural networks are commonly trained by the traditional, back-propagation learning format, which minimises the mean squared error (mse) of the training data. However, in the presence of corrupted or extremely deviated samples (outliers), this training scheme may produce incorrect models, and it is well known that industrial data sets frequently contain outliers. The process modelled is a steel coil annealing furnace in a galvanising line, which shares characteristics with most of the furnaces used in galvanised lines all over the world. This paper reports the effectiveness of robust learning algorithms compared to the classical mse-based learning algorithm for the modelling of a real industry process. From this model an adequate line velocity (the velocity set point) for a coil, depending on its characteristics and the furnace condition to receive this coil (temperature set points), can be obtained. With this set point generation model the operator could set strategies to manage the line, i.e. set the order of the coil to be treated or preview the line's speed conditions for the transitory situations.     

A new procedure for improving the predictiveness of CoMFA models and its application to a set of dihydrofolate reductase inhibitors

A new automated procedure to improve the predictive quality of CoMFA models for both training and test sets is described. A model of greater consistency is generated by performing small reorientations of the underlying molecules for which too low activities are calculated. In order to predict activities of test compounds, the most similar molecules in the previously optimized model are identified and used as a basis for the prediction. This method has been applied to two independent sets of dihydrofolate reductase inhibitors (80 compounds each, serving as training sets), resulting in a significant increase of the cross-validated r2 value. For both models, the predictive r2 value for a test set consisting of 70 compounds was improved substantially.     

Some applications of neural networks for prediction of blast furnace irregularities

The on-line analysis of operational data and prediction of furnace irregularities, though difficult, are essential for the improvement of the control of blast furnace operation. Three models based on artificial neural networks for the recognition of top gas distribution, distributions of the heat fluxes through the furnace wall, and for the prediction of slips have been designed. The off-line test results showed that a trained perceptron network could recognise various types of top gas profiles. A classifier consisting of a self-organising feature map network and a learning vector quantizer could classify the characteristic patterns of heat flux distribution; and a model based on a back propagation network could properly predict the probability of upcoming slips in advance. The most important operational variables needed for predicting slips have also been extracted. It has been proved that the neural network used has a good capability of predicting furnace irregularities.