Neural Network-based Methodology Research Articles

Deep-Hipo: Multi-scale receptive field deep learning for histopathological image analysis.

Digitizing whole-slide imaging in digital pathology has led to the advancement of computer-aided tissue examination using machine learning techniques, especially convolutional neural networks. A number of convolutional neural network-based methodologies have been proposed to accurately analyze histopathological images for cancer detection, risk prediction, and cancer subtype classification. Most existing methods have conducted patch-based examinations, due to the extremely large size of histopathological images. However, patches of a small window often do not contain sufficient information or patterns for the tasks of interest. It corresponds that pathologists also examine tissues at various magnification levels, while checking complex morphological patterns in a microscope. We propose a novel multi-task based deep learning model for HIstoPatholOgy (named Deep-Hipo) that takes multi-scale patches simultaneously for accurate histopathological image analysis. Deep-Hipo extracts two patches of the same size in both high and low magnification levels, and captures complex morphological patterns in both large and small receptive fields of a whole-slide image. Deep-Hipo has outperformed the current state-of-the-art deep learning methods. We assessed the proposed method in various types of whole-slide images of the stomach: well-differentiated, moderately-differentiated, and poorly-differentiated adenocarcinoma; poorly cohesive carcinoma, including signet-ring cell features; and normal gastric mucosa. The optimally trained model was also applied to histopathological images of The Cancer Genome Atlas (TCGA), Stomach Adenocarcinoma (TCGA-STAD) and TCGA Colon Adenocarcinoma (TCGA-COAD), which show similar pathological patterns with gastric carcinoma, and the experimental results were clinically verified by a pathologist. The source code of Deep-Hipo is publicly available athttp://dataxlab.org/deep-hipo.

Evaluation of neural network-based methodologies for wind speed forecasting

Multi-layer perceptron neural networks (MLP-NN) are widely utilized in forecasting applications. However, optimal training of these networks is still a challenge. A comprehensive assessment of MLP training approaches comprising of three stages is performed in the present paper. First, the prediction performance is evaluated using twelve training algorithms. Next, optimization algorithms are utilized to enhance the best obtained network parameters obtained from the first step and the performance of eight optimization algorithms is evaluated. Finally, a novel modification is used to improve the performance of the optimization algorithms. The proposed methodologies are applied to two case-studies and statistical metrics are employed for their efficiency evaluation. Wavelet transformation is used to extract the features which will be fed to the MLP-NN as input data.

A computational Intelligence-based Method to ‘Learn’ Causal Loop Diagram-like Structures from Observed Data

The development of conceptual models using causal loop diagrams and their variants is a key step in the system dynamics modeling process. This work seeks to explore to what extent such models can be inferred directly from system observations using computational methods. A novel echo state neural network-based methodology is proposed to automatically learn causal loop diagram-like structures directly from system observations. The proposed data-driven approach aims at complementing the conceptual model development process by providing modelers with several probable model structures that can be accepted readily or considered for refinement. Three measures, used in comparing mental models, are adopted to compute similarity between the learned and target model structures. Using three well-known system dynamics case studies, we show the effectiveness of the proposed method in learning close model structures directly from the system observations, generated by simulating the stock-and-flow models for these cases.Copyright © 2017 System Dynamics Society

Accuracy and Calibration of Computational Approaches for Inpatient Mortality Predictive Modeling.

Electronic Health Record (EHR) data can be a key resource for decision-making support in clinical practice in the “big data” era. The complete database from early 2012 to late 2015 involving hospital admissions to Inselspital Bern, the largest Swiss University Hospital, was used in this study, involving over 100,000 admissions. Age, sex, and initial laboratory test results were the features/variables of interest for each admission, the outcome being inpatient mortality. Computational decision support systems were utilized for the calculation of the risk of inpatient mortality. We assessed the recently proposed Acute Laboratory Risk of Mortality Score (ALaRMS) model, and further built generalized linear models, generalized estimating equations, artificial neural networks, and decision tree systems for the predictive modeling of the risk of inpatient mortality. The Area Under the ROC Curve (AUC) for ALaRMS marginally corresponded to the anticipated accuracy (AUC = 0.858). Penalized logistic regression methodology provided a better result (AUC = 0.872). Decision tree and neural network-based methodology provided even higher predictive performance (up to AUC = 0.912 and 0.906, respectively). Additionally, decision tree-based methods can efficiently handle Electronic Health Record (EHR) data that have a significant amount of missing records (in up to >50% of the studied features) eliminating the need for imputation in order to have complete data. In conclusion, we show that statistical learning methodology can provide superior predictive performance in comparison to existing methods and can also be production ready. Statistical modeling procedures provided unbiased, well-calibrated models that can be efficient decision support tools for predicting inpatient mortality and assigning preventive measures.

A neural network approach for identifying particle pitch angle distributions in Van Allen Probes data

Analysis of particle pitch angle distributions (PADs) has been used as a means to comprehend a multitude of different physical mechanisms that lead to flux variations in the Van Allen belts and also to particle precipitation into the upper atmosphere. In this work we developed a neural network-based data clustering methodology that automatically identifies distinct PAD types in an unsupervised way using particle flux data. One can promptly identify and locate three well-known PAD types in both time and radial distance, namely, 90deg peaked, butterfly, and flattop distributions. In order to illustrate the applicability of our methodology, we used relativistic electron flux data from the whole month of November 2014, acquired from the Relativistic Electron-Proton Telescope instrument on board the Van Allen Probes, but it is emphasized that our approach can also be used with multiplatform spacecraft data. Our PAD classification results are in reasonably good agreement with those obtained by standard statistical fitting algorithms. The proposed methodology has a potential use for Van Allen belt's monitoring.

A neural network-based methodology for inverse kinematics of a multi-finger robotic hand for gripping

A neural network-based methodology for inverse kinematics of a multi-finger robotic hand for gripping

A neural network-based methodology for inverse kinematics of a multi-finger robotic hand for gripping

Robotic grasping and manipulation require controlling the gripper movement through different points in its work volume, necessitating inverse kinematics computations to determine joint angles. In the present work, a novel methodology, based on a radial basis function neural network, has been proposed for the inverse kinematics solution and a genetic algorithm-based approach for optimising the neural network parameters. Instead of taking the entire work volume of the hand for neural network training, a subspace of points is created in close vicinity of the given destination point. The joint variables corresponding to a destination point are obtained using a random walk algorithm that uses the forward kinematics model of the hand. Then, the subspace of points and the corresponding joint variables obtained above are used to train the neural network. This approach can provide an approximate yet fairly quick and effective solution to the inverse kinematics problem of multi-finger robot hands.

A new technique for lithology and fluid content prediction from prestack data: An application to a carbonate reservoir

One of the leading challenges in hydrocarbon recovery is predicting rock types/fluid content distribution throughout the reservoir away from the boreholes because rock property determination is a major source of uncertainty in reservoir modeling studies. Spatial determination of the lateral and vertical heterogeneities has a direct impact on a reservoir model because it will affect the property distributions. An inappropriate determination of the facies distribution will lead to unrealistic reservoir behavior. Because these data can take different forms (lithologs, cuttings, and for seismic, poststack, and prestack attributes) and have different resolutions, the manual integration of all the information can be tedious and is sometimes impractical. We developed a new neural network-based methodology called democratic neural network association (DNNA). The DNNA method was trained using lithology logs from wells simultaneously with prestack seismic data. This technique, using a probabilistic approach, aims to find patterns in seismic that will predict lithology distribution and uncertainty.

A neural network-based methodology of quantifying the association between the design variables and the users’ performances

User performance is highly correlated with design variables of a system. Such association can be described as display–control relationship. In this study, a neural network-based methodology is proposed to identify and quantify the association among design variables (inputs) and to compute their relative influences (RIs) on the two performance measures (outputs) of user response time and response accuracy, using artificial neural network, generalised regression neural network, support vector regression (SVR), multiple linear regression and response surface model. Based on the results of the comparison, it is found that neural network-based methods are more reliable than SVR-based methods in computing the RI of design variables. As a result of our analysis, the best option for optimising each of the measures is suggested. Some useful observations about the design of man–machine systems are also presented, discussed and visualised. In the study of man–machine systems, quantitative methods are seldom adopted for examining the mappings between various displays and controls under a variety of operating conditions. The major contribution of this study is to provide some insights into the usefulness of quantitative methods in evaluating man–machine design in terms of display–control compatibility and to extract explanatory information from renowned black box systems such as neural networks.

ANN-based Methodology to Determine Dynamic to Static Eccentricity Ratio of Torsionally Coupled Buildings for Site-Specific Earthquakes

Although excessive damage of torsionally coupled buildings situated on soil sites has been reported during past earthquakes, no site-specific study is available having depth of soil stratum above rock as a parameter. In this article, an Artificial Neural Network-based methodology to determine dynamic to static eccentricity ratio of torsionally coupled buildings is proposed. Twelve neural networks are developed using 648,000 patterns from the analyses of shear type single-story building models. The methodology is found to be appropriate to determine dynamic to static eccentricity ratio for multi-story uniform, non uniform framed, and shear wall buildings for site-specific earthquake.

Influence of caffeine and temperature on corrosion-resistance of CoCrMo alloy

AbstractThe inhibitory activity of caffeine (1,3,7-trimethyl xanthine) on artificial saliva was studied on a CoCrMo alloy using different electrochemical methods: open circuit potential (OCP), potentiodynamic measurements and electrochemical impedance spectroscopy (EIS). The results show that caffeine produces an inhibitory effect on the anodic currents due to its adsorption on the surface of the alloy. Temperature is another parameter with an influence on corrosion processes, so thermodynamic data were obtained from Arrhenius plots and Langmuir adsorption isotherms. The protective action of caffeine is enhanced at high temperatures at OCP, while for potentiodynamic experiments high temperatures block the inhibitory activity of caffeine and the corrosion rate increases. The process may also be studied by a simulation, determining the functional dependence between OCP, corrosion current density (i corr), corrosion potential (E corr), breakdown potential (E bd) and temperature and amount of caffeine in artificial saliva, for Heraenium® CE. The neural network-based methodology applied in this work provides accurate results, thus proving to be an efficient modelling technique.

Neural Network Model of Estimation of Body Mass Index Based on Indirect Input Factors

A well-prepared One of the main concerns of people in developing and developed societies is increasing the Body Mass Index (BMI) level. BMI, in fact can be considered as an indicator of overall health condition. Genetic aspects aside, the BMI level is affected by different factors, such as socio-economic, environmental, and physical activity level. This study investigated the effect of different factors on the BMI level of a sample population of 470 adults of three residential neighborhoods in Shiraz, Iran. The Pearson correlation coefficient, independent sample T-test and One Way ANOVA were used to extract the variables which significantly influenced the BMI. The statistical analysis showed that despite the apparent association of BMI with physical activity level, it is influenced by several factors such as age, residence record, number of children, distance to bus or taxi stop, indoor or sport exercise. Then, an Artificial Neural Network (ANN) was applied to predict the level of personal BMI. Artificial Neural Network-based methodology results showed that the generalized estimating ANN model was satisfactory in estimating the BMI based on the introduced pattern.

Neural network-based methodology for inter-arrival times of earthquakes

In this paper, an artificial neural network (ANN)–based methodology is proposed to determine the probability of inter-arrival time (IAT) of main shock of six broad seismic regions of India. Initially, classical methodology using exponential distribution is applied to IAT of earthquake events computed from earthquake catalog data. From the goodness-of-fit test results, it has been found that exponential distribution is not adequate. In this paper, a more efficient ANN-based methodology is proposed, and two ANN models are developed to determine the probability of IAT of earthquake events for a specified region, specified magnitude range or magnitude greater than the specified value. The performance of ANN models developed is validated with number of examples and found to predict the probability with minimal error compared to exponential distribution model. The methodology developed can be applied to any other region with the database of the respective regions.

Designing a Visual Inspection System for Quality Characteristics Dimensions

Industrial inspection is one of the crucial tasks to ensure quality conformance of products. The inspection tasks can be done by using several methods like non-scaled go/not go gauging, measuring instruments, or advanced non-touching tools. In this research visual inspection using a developed optical system is conducted. One of the aims of this research is to design an on line visual inspection system that is capable to test geometrical quality characteristics of 2-D machined products. The design process includes developing an economical optical system to acquire inspected product’s images. Image processing tools are utilized to deal with the product image; and extract features of its geometrical characteristics. A neural network-based methodology is developed and applied to decide whether the product conforms to pre-specified tolerances. The results of the developed methodology are compared to some statistics based visual approaches from the literature. The results show the goodness of the system as an automated visual inspection system and prove its superior performance with respect to other methods.

Neural Network Classification of Brain Hemodynamic Responses from Four Mental Tasks

We investigate subjects’ brain hemodynamic activities during mental tasks using a nearinfrared spectroscopy. A wavelet and neural network-based methodology is presented for recognition of brain hemodynamic responses. The recognition is performed by a single layer neural network classifier according to a backpropagation algorithm with two error minimizing techniques. The performance of the classifier varied depending on the neural network model, but the performance was usually at least 90%. The classifier usually converged faster and attained a somewhat greater level of performance when an input was presented with only relevant features. The overall classification rate was higher than 94%. The study demonstrates the accurate classifiablity of human brain hemodynamic useful in various brain studies.

Path Characteristic Frequency-Based Fault Locating in Radial Distribution Systems Using Wavelets and Neural Networks

This paper proposes a new algorithm for localizing phase to ground faults in an electric power distribution system. Fault-originated traveling waves propagate along the distribution paths in both directions away from the fault point and are reflected at line terminations, junctions between feeders, laterals, and the fault location. Depending on the paths which traveling waves reciprocate through, the transient signal at each node contains certain path characteristic frequencies (PCFs). Energy spectrum of the transient signal has high density around the path characteristic frequencies. On this basis, the transient voltage is decomposed by a wavelet filter and its energy spectrum is decomposed into different levels. Depending on the bandwidth of the wavelet filter and the path characteristic frequencies, the decomposed signal in each level contains a certain percentage of energy. Then, a neural network-based methodology is proposed as a power distribution line fault locator. Energy percentage in each level is the candidate feature for training the artificial neural network. Simulations and the training process for the neural network are performed respectively using ATP/EMTP and MATLAB. It is shown that the results of the proposed algorithm are quite satisfactory.

Load Estimation in Balanced Radial Distribution System with Reduced Measurements

This article presents a technique for real-time load estimation in a balanced radial power distribution system in which the number of measurements required is significantly less than the number of load quantities to be estimated. As the number of measurements is lower than the number of quantities to be estimated, the weighted least square method of estimation cannot be used, and therefore, an artificial neural network-based methodology is used for estimating the loads. From extensive performance evaluation studies on different distribution systems, it was found that with the meters currently available in the market, an estimation error of approximately 1% can be obtained with the proposed methodology.

Analysis of University Science Research Capability Elements and Evaluation Based on BP Neural Network

After analyzing effect elements of University Science Research Capability (USRC) based on dynamic capability theory, combined the substance of university science research with the highly self-organized, self-adapted and self-learned characteristics of Back Propagation (BP) Neural Network, the paper conducts a research on evaluation of USRC, in which an evaluation index system of USRC is constructed and a 15-7-1-typed BP Neural Network with three layers is presented to evaluate USRC, which provides a BP Neural Network-based methodology for evaluation of USRC with multiple inputs.

Weed and crop discrimination using image analysis and artificial intelligence methods

Development of a visual method of discriminating between crop seedlings and weeds is an important and necessary step towards the automation of non-chemical weed control systems in agriculture, and towards the reduction in chemical use through spot spraying. Two methods were applied to recognise carrot ( Daucus carota L.) seedlings from those of ryegrass ( Lolium perenne) and Fat Hen ( Chenopodium album) using digital imaging. The first method involved the use of a simple morphological characteristic measurement of leaf shape (perimeter 2/area), which had varying effectiveness (between 52 and 74%) in discriminating between the two types of plant, with the variation dependent on plant size. The second involved a self-organising neural network more biologically plausible than many commonly used NN methods. While the latter did not give results as good as those required for commercial purposes, it showed that a neural network-based methodology exists which allows the system to learn and discriminate between species to an accuracy exceeding 75% without predefined plant descriptions being necessary.

Extraction and use of neural network models in automated synthesis of operational amplifiers

Fast and accurate performance estimation methods are essential to automated synthesis of analog circuits. Development of analog performance models is difficult due to the highly nonlinear nature of various analog performance parameters. This paper presents a neural network-based methodology for creating fast and efficient models for estimating the performance parameters of CMOS operational amplifier topologies. Effective methods for generation and use of the training data are proposed to enhance the accuracy of the neural models. The efficiency and accuracy of the resulting performance models are demonstrated via their use in a genetic algorithm-based circuit synthesis system. The genetic synthesis tool optimizes a fitness function based on user-specified performance constraints. The performance parameters of the synthesized circuits are validated by SPICE simulations and compared with those predicted by the neural network models. Experimental studies demonstrate that neural network modeling is an effective, fast, and accurate methodology for performance estimation.