Stuttgart Neural Network Simulator Research Articles

Machine Learning Models of Survival Prediction in Trauma Patients.

Background: We aimed to build a model using machine learning for the prediction of survival in trauma patients and compared these model predictions to those predicted by the most commonly used algorithm, the Trauma and Injury Severity Score (TRISS). Methods: Enrolled hospitalized trauma patients from 2009 to 2016 were divided into a training dataset (70% of the original data set) for generation of a plausible model under supervised classification, and a test dataset (30% of the original data set) to test the performance of the model. The training and test datasets comprised 13,208 (12,871 survival and 337 mortality) and 5603 (5473 survival and 130 mortality) patients, respectively. With the provision of additional information such as pre-existing comorbidity status or laboratory data, logistic regression (LR), support vector machine (SVM), and neural network (NN) (with the Stuttgart Neural Network Simulator (RSNNS)) were used to build models of survival prediction and compared to the predictive performance of TRISS. Predictive performance was evaluated by accuracy, sensitivity, and specificity, as well as by area under the curve (AUC) measures of receiver operating characteristic curves. Results: In the validation dataset, NN and the TRISS presented the highest score (82.0%) for balanced accuracy, followed by SVM (75.2%) and LR (71.8%) models. In the test dataset, NN had the highest balanced accuracy (75.1%), followed by the TRISS (70.2%), SVM (70.6%), and LR (68.9%) models. All four models (LR, SVM, NN, and TRISS) exhibited a high accuracy of more than 97.5% and a sensitivity of more than 98.6%. However, NN exhibited the highest specificity (51.5%), followed by the TRISS (41.5%), SVM (40.8%), and LR (38.5%) models. Conclusions: These four models (LR, SVM, NN, and TRISS) exhibited a similar high accuracy and sensitivity in predicting the survival of the trauma patients. In the test dataset, the NN model had the highest balanced accuracy and predictive specificity.

ANN based methodology for active control of buildings for seismic excitation for different seismic zones of India

Abstract In the last two decades, many studies are reported in literature for determining the control force for the active control systems for damage mitigation of buildings due to earthquake. However, no study has been reported for prediction of control force considering the seismic zones in India. In the present study, spectrum compatible time histories are generated for the seismic zones IV and V as per Indian standard IS 1893(Part 1):2002 design spectrum. Time history analysis are carried out with spectrum compatible time histories for shear type buildings modelled as multi-degree of freedom system (MDOF) with a computer program developed in MATLAB by modal superposition using Newmark-beta method. Control forces are obtained by adopting algorithm proposed in literature and input and output patterns are generated for development of Artificial Neural Network (ANN) models in Stuttgart Neural Network Simulator (SNNS). In the present study the methodology is demonstrated with five storey building by developing 24 ANN models consisting of two ANN architectures viz., NET1 and NET2 for each seismic zone and three soil types. From the validation of results from ANN models it is observed that the maximum difference in percentage response reduction of peak displacement is less than 10% when it is compared with the target value of percentage response reduction.

Investigating the performance of neural network backpropagation algorithms for TEC estimations using South African GPS data

Abstract. In this work, results obtained by investigating the application of different neural network backpropagation training algorithms are presented. This was done to assess the performance accuracy of each training algorithm in total electron content (TEC) estimations using identical datasets in models development and verification processes. Investigated training algorithms are standard backpropagation (SBP), backpropagation with weight delay (BPWD), backpropagation with momentum (BPM) term, backpropagation with chunkwise weight update (BPC) and backpropagation for batch (BPB) training. These five algorithms are inbuilt functions within the Stuttgart Neural Network Simulator (SNNS) and the main objective was to find out the training algorithm that generates the minimum error between the TEC derived from Global Positioning System (GPS) observations and the modelled TEC data. Another investigated algorithm is the MatLab based Levenberg-Marquardt backpropagation (L-MBP), which achieves convergence after the least number of iterations during training. In this paper, neural network (NN) models were developed using hourly TEC data (for 8 years: 2000–2007) derived from GPS observations over a receiver station located at Sutherland (SUTH) (32.38° S, 20.81° E), South Africa. Verification of the NN models for all algorithms considered was performed on both "seen" and "unseen" data. Hourly TEC values over SUTH for 2003 formed the "seen" dataset. The "unseen" dataset consisted of hourly TEC data for 2002 and 2008 over Cape Town (CPTN) (33.95° S, 18.47° E) and SUTH, respectively. The models' verification showed that all algorithms investigated provide comparable results statistically, but differ significantly in terms of time required to achieve convergence during input-output data training/learning. This paper therefore provides a guide to neural network users for choosing appropriate algorithms based on the availability of computation capabilities used for research.

Neural Networks inRUsing the Stuttgart Neural Network Simulator:RSNNS

Neural networks are important standard machine learning procedures for classication and regression. We describe the R package RSNNS that provides a convenient interface to the popular Stuttgart Neural Network Simulator SNNS. The main features are (a) encapsulation of the relevant SNNS parts in a C++ class, for sequential and parallel usage of dierent networks, (b) accessibility of all of the SNNS algorithmic functionality from R using a low-level interface, and (c) a high-level interface for convenient, R-style usage of many standard neural network procedures. The package also includes functions for visualization and analysis of the models and the training procedures, as well as functions for data input/output from/to the original SNNS le formats.

LavaNet—Neural network development environment in a general mine planning package

LavaNet is a series of scripts written in Perl that gives access to a neural network simulation environment inside a general mine planning package. A well known and a very popular neural network development environment, the Stuttgart Neural Network Simulator, is used as the base for the development of neural networks. LavaNet runs inside VULCAN™—a complete mine planning package with advanced database, modelling and visualisation capabilities. LavaNet is taking advantage of VULCAN's Perl based scripting environment, Lava, to bring all the benefits of neural network development and application to geologists, mining engineers and other users of the specific mine planning package. LavaNet enables easy development of neural network training data sets using information from any of the data and model structures available, such as block models and drillhole databases. Neural networks can be trained inside VULCAN™ and the results be used to generate new models that can be visualised in 3D. Direct comparison of developed neural network models with conventional and geostatistical techniques is now possible within the same mine planning software package. LavaNet supports Radial Basis Function networks, Multi-Layer Perceptrons and Self-Organised Maps.

Correction for crosstalk contaminations in dual radionuclide /sup 99m/Tc and /sup 123/I images using artificial neural network

Use of an artificial neural network (ANN) has been previously shown to be an effective tool in compensating scatter and crosstalk from the primary photons in simultaneous dual radionuclide imaging. Generally, a large number of input energy windows are required within the network structure while the commercial cameras have only 3-8 energy windows. It is difficult to use two input windows within the ANN structure for the crosstalk contamination corrections of /sup 99m/Tc//sup 123/I images acquired using only two photopeak energy windows. In this paper, we designed an ANN network with 24 inputs, 32 nodes in the hidden layer and two nodes in the output layer, to correct for crosstalk contamination in /sup 99m/Tc//sup 123/I images acquired using two photopeak windows. We trained the network using experimentally acquired /sup 99m/Tc and /sup 123/I spectrum data using the RSD brain phantom. The neural network package Stuttgart Neural Network Simulator (SNNSv4.2), from the University of Stuttgart, was used for the neural network training and the crosstalk corrections. Two sets of image data were tested. The first was a human activation study and the other used a cylindrical striatal phantom. Our results show a great improvement on both the human activation and the cylindrical striatal phantom images. Further work is to test our new approach on more /sup 99m/Tc//sup 123/I imaging data and apply it to other radionuclide combinations such as /sup 201/Tl//sup 99m/Tc.

An Optical-Fiber Sensor for Use in Water Systems Utilizing Digital Signal Processing Techniques and Artificial Neural Network Pattern Recognition

An optical-fiber sensor is reported which is capable of detecting ethanol in water. A single optical-fiber sensor was incorporated into a 1-km length of 62.5-/spl mu/m core diameter polymer-clad silica optical fiber. In order to maximize sensitivity, a U-bend configuration was used for the sensor where the cladding was removed and the core exposed directly to the fluid under test. The sensor was interrogated using optical time domain reflectrometry, as it is intended to extend this work to multiple sensors on a single fiber. In this investigation, the sensor was exposed to air, water, and alcohol. The signal processing technique has been designed to optimize the neural network adopted in the existing sensor system. In this investigation, a discrete Fourier transform, using a fast Fourier transform algorithm, is chosen and its application leads to an improvement in efficiency of the neural network i.e., minimizing the computing resources. Using the Stuttgart neural network simulator, a feed-forward three-layer neural network was constructed with the number of input nodes corresponding to the number of points required to represent the sensor frequency domain response.

An optical fibre ethanol concentration sensor utilizing Fourier transform signal processing analysis and artificial neural network pattern recognition

An optical fibre sensor, which is capable of detecting varying percentages of ethanol in water, is reported. In order to maximize sensitivity, a U-bend configuration was used for the sensor where the cladding was removed and the core exposed directly to the fluid under test. The sensor was interrogated using optical time domain reflectometry (OTDR). OTDR is chosen as it is a recognized technique for the interrogation of distributed multipoint sensors and it is intended to extend this work to multiple sensors on a single fibre in the future. In this investigation the sensor was exposed to 12.5, 25 and 50% ethanol and distilled water. The signal processing technique has been designed to optimize the neural network adopted in the existing sensor system. In this investigation a discrete Fourier transform, using a fast Fourier transform algorithm, is chosen and its application leads to an improvement in efficiency of the neural network, i.e. reducing the number of input and hidden layer nodes required by the artificial neural network. Using a Stuttgart neural network simulator, a feed-forward three-layer neural network was constructed with the aim of successfully classifying the sensor test conditions based on the frequency domain response of the sensor.

Prediction of Earth orientation parameters by artificial neural networks

Earth orientation parameters (EOPs) [polar motion and length of day (LOD), or UT1–UTC] were predicted by artificial neural networks. EOP series from various sources, e.g. the C04 series from the International Earth Rotation Service and the re-analysis optical astrometry series based on the HIPPARCOS frame, served for training the neural network for both short-term and long-term predictions. At first, all effects which can be described by functional models, e.g. effects of the solid Earth tides and the ocean tides or seasonal atmospheric variations of the EOPs, were removed. Only the differences between the modeled and the observed EOPs, i.e. the quasi-periodic and irregular variations, were used for training and prediction. The Stuttgart neural network simulator, which is a very powerful software tool developed at the University of Stuttgart, was applied to construct and to validate different types of neural networks in order to find the optimal topology of the net, the most economical learning algorithm and the best procedure to feed the net with data patterns. The results of the prediction were analyzed and compared with those obtained by other methods. The accuracy of the prediction is equal to or even better than that by other prediction methods.

Evolution of neural networks for radiative transfer calculations in the terrestrial infrared

Land surface temperature (LST) is an important component of the energy-budget of the surface. In order to determine the brightness temperature (BT) measured by a satellite for a given LST, the atmospheric influence on radiance emitted by the surface has to be accounted for. Provided that the current state of the atmosphere (vertical moisture and temperature profiles) and the surface emissivity are sufficiently well known, it is possible to calculate the BT using a radiative transfer model (RTM), e.g., Moderate Resolution Transmittance Code 3 (MODTRAN-3). RTMs do not linearise the atmospheric effect and variations of surface emissivity, elevation, and view angle (i.e., the path through the atmosphere) are readily accounted for. However, RTMs are very expensive in terms of computing time and, therefore, not well suited to simulate large quantities of data. In order to overcome this limitation, the main objective of this study is to investigate if MODTRAN-3 can be substituted by a feed-forward neural network (NN). Training and validation data consist of inputs and outputs of MODTRAN-3 for 84 TOVS Initial Guess Retrieval (TIGR) profiles in the middle latitudes (45°N to 55°N). The RTM was run for these atmospheric situations and a range of LSTs, surface elevations, and view angles. In total, 4032 radiative transfer calculations were performed. A NN developed manually for this data is evaluated using forward calculations for atmospheric profiles from the European Centre of Medium-Range Weather Forecasts (ECMWF) analyses. Grid cells of the ECMWF analyses containing clouds are identified using cloudmasks derived from Meteosat IR and VIS data. The manually developed NN is compared against a NN developed by the evolutionary algorithm “Evolutionärer Netzwerk Optimierer” (ENZO). ENZO evolves populations of NNs: it utilises mutation and crossover to optimise NN topology (layers, number of neurons, and weights) and trains (fine-tunes) the NNs with the Stuttgart Neural Network Simulator (SNNS). The NN developed by ENZO has a rms validation error (untrained TIGR situations) of 0.25 K and a rms verification error (cloud-free situations from ECMWF analyses) of 0.31 K. ENZO is shown to produce substantially smaller NNs than the manual approach: the developed NN has 58% fewer hidden neurons and about 90% fewer weights than the manually developed NN. Furthermore, the NN developed by ENZO is more robust and generalises better. A major advantage of the NN is its computational speed, which is estimated to be of the order of 10 4 times faster than MODTRAN-3.

Application of Neural Networks for unfolding neutron spectra measured by means of Bonner Spheres

A Neural Network structure has been used for unfolding neutron spectra measured by means of a Bonner Sphere Spectrometer set. The present work used the “Stuttgart Neural Network Simulator” as the interface for designing, training and validation of a MultiLayer Perceptron network. The back-propagation algorithm was applied. The Bonner Sphere set chosen has been calibrated at the National Physical Laboratory, United Kingdom, and uses gold activation foils as thermal neutron detectors. The neutron energy covered by the response functions goes from 0.0001 eV to 10 MeV. Two types of neutron spectra were numerically investigated: monoenergetic and continuous. Good results were obtained, indicating that the Neural Network can be considered an interesting alternative among the neutron spectrum unfolding methodologies.

Neural network-based estimation of stress concentration factors for steel multiplanar tubular XT-joints

The hot-spot stress method for fatigue design of tubular joints relies on the accurate predictions of the stress concentration factors (SCF) at the brace to chord intersection areas. At present, SCFs are predicted based on established empirical equations. An alternative approach using a neural network-based model has been developed in this paper to estimate the SCFs of multiplanar tubular XT-joints. The neural network software, Stuttgart Neural Network Simulator, was used for the purpose. To train and test the network, an SCF database was built up using the finite element method (FEM). The database covers a wide range of geometrical parameters for the XT-joints. Three axial load cases were considered. The geometrical properties of the tubular joints were used as the training input data. The FEM SCFs are used as the training output data. Different network configurations are also tested for the best possible results. The results show that a trained neural network can indeed predict the SCFs for the various load cases with a higher level of accuracy.

Stuttgart Neural Network Simulator: Exploring connectionism and machine learning with SNNS

Stuttgart Neural Network Simulator: Exploring connectionism and machine learning with SNNS