Unsupervised Artificial Neural Network Research Articles

Unsupervised extraction of urban features from airborne lidar data by using self-organizing maps

The extraction of artificial and natural features using light detection and ranging (Lidar) data is a fundamental task in many fields of research for environmental science. In this study, the possibility of using self-organising maps (SOM), which is an unsupervised artificial neural network classification method to extract the bare earth surface and features from airborne Lidar data, was investigated for two different urban areas. The effect of the enlargement of the study area was analysed using the proposed approach. The appropriate weights of SOM inputs, which are 3D coordinates and intensity, obtained from a Lidar point cloud were determined by using Pearson's chi-squared independence test. The weighted SOM feature extraction performance was better than that of the unweighted SOM. The filtering results of SOM to separate ground and non-ground data were also compared with those obtained by the adaptive TIN filtering algorithm. Most of the non-ground features could be removed by the weighted SOM.

Prediction of reservoir quality using well logs and seismic attributes analysis with an artificial neural network: A case study from Farrud Reservoir, Al-Ghani Field, Libya

This paper presents an innovative technique that aims to predict the quality of a petroleum reservoir using Artificial Neural Networks (ANN) analysis of seismic, well logs and core data. A supervised Probabilistic Neural Network (PNN) was deployed to predict several reservoir properties, one at a time, through training and validation of the PNN to determine the seismic attributes that best fit a measured property in the well logs. The validated PPN models accurately converted the available 3D seismic data into shale volume, porosity, permeability and water saturation cubes. In addition, these predicted reservoir properties were integrated to define various reservoir grades using K-means Clustering algorithm of unsupervised classification ANN. This technique is applied to Al-Ghani Field and four grades of reservoir quality were classified as very good, good, bad, and very bad and their spatial distributions were displayed. The highest reservoir grade is characterized by good porosity and permeability and significantly low water saturation. Such information is highly valuable for optimum reservoir management and well placement. This not only maximizes reservoir profitability through production schemes, but also minimizes uncertainties in drilling, production, injection, and modeling processes. In addition, adopting the proposed methodology would decrease costs in well logging programs and improve the sweeping efficiency of water flooding operations.

A new stochastic computing paradigm for the dynamics of nonlinear singular heat conduction model of the human head

Bio-inspired computing approaches are effective to solve a variety of dynamical problems. The strength of these stochastic solvers is exploited for the numerical treatment of a nonlinear heat conduction model of the human head using artificial neural networks (ANNs), genetic algorithms (GAs), active-set technique (AST), and their hybrids. The universal function approximation competencies of unsupervised ANNs are utilized in constructing the mathematical model of the problem by defining an error function in the mean squared sense. The training of the design parameters of ANN models is made with global search brilliance of GAs, viably local search with AST and hybrid approach GA-AST. The results of the proposed schemes are determined in terms of temperature profiles by considering variants of the problem with different Biot numbers, metabolic thermogenesis slope parameters and thermogenesis heat production factors. The correctness, effectiveness and convergence of the proposed approaches are also ascertained through statistics.

Enhancing Performance of Parallel Self-Organizing Map on Large Dataset with Dynamic Parallel and Hyper-Q

Self-Organizing Map (SOM) is an unsupervised artificial neural network algorithm. Even though this algorithm is known to be an appealing clustering method,many efforts to improve its performance are still pursued in various research works. In order to gain faster computation time, for instance, running SOM in parallel had been focused in many previous research works. Utilization of the Graphics Processing Unit (GPU) as a parallel calculation engine is also continuously improved. However, total computation time in parallel SOM is still not optimal on processing large dataset. In this research, we propose a combination of Dynamic Parallel and Hyper-Q to further improve the performance of parallel SOM in terms of faster computing time. Dynamic Parallel and Hyper-Q are utilized on the process of calculating distance and searching best-matching unit (BMU), while updating weight and its neighbors are performed using Hyper-Q only. Result of this study indicates an increase in SOM parallel performance up to two times faster compared to those without using Dynamic Parallel and Hyper-Q.

Optimization of sediment rating curve coefficients using evolutionary algorithms and unsupervised artificial neural network

Sediment rating curve (SRC) is a conventional and a common regression model in estimating suspended sediment load (SSL) of flow discharge. However, in most cases the data log-transformation in SRC models causing a bias which underestimates SSL prediction. In this study, using the daily stream flow and suspended sediment load data from Shalman hydrometric station on Shalmanroud River, Guilan Province, Iran, SRC equation was derived, and then, using evolutionary algorithms (genetic algorithm and particle swarm optimization algorithm) it was calibrated again. Worth mentioning, before model calibration, to increase the generalization power of the models, using self-organizing map (an unsupervised artificial neural network for data clustering), the data were clustered and then by data sampling, they were classified into two homogeneous groups (calibration and test data set). The results showed that evolutionary algorithms are appropriate methods for optimizing coefficients of SRC model and their results are much more favorable than those of the conventional SRC models or SRC models corrected by correction factors. So that, the sediment rating curve models calibrated with evolutionary algorithms, by reducing the RMSE of the test data set of 5754.02 ton day-1 (in the initial SRC model) to 1681.21 ton day-1 (in the calibrated models by evolutionary algorithms) increased the accuracy of suspended sediment load estimation at a rate of 4072.81 ton day-1. In total, using evolutionary algorithms in calibrating SRC models prevents data log-transformation and use of correction factors along with increasing in the accuracy of molding results.

Machine learning in pain research.

Machine learning in pain research.

Semi-automatic mapping of linear-trending bedforms using ‘Self-Organizing Maps’ algorithm

Increased application of high resolution spatial data such as high resolution satellite or Unmanned Aerial Vehicle (UAV) images from Earth, as well as High Resolution Imaging Science Experiment (HiRISE) images from Mars, makes it necessary to increase automation techniques capable of extracting detailed geomorphologic elements from such large data sets. Model validation by repeated images in environmental management studies such as climate-related changes as well as increasing access to high-resolution satellite images underline the demand for detailed automatic image-processing techniques in remote sensing. This study presents a methodology based on an unsupervised Artificial Neural Network (ANN) algorithm, known as Self Organizing Maps (SOM), to achieve the semi-automatic extraction of linear features with small footprints on satellite images. SOM is based on competitive learning and is efficient for handling huge data sets. We applied the SOM algorithm to high resolution satellite images of Earth and Mars (Quickbird, Worldview and HiRISE) in order to facilitate and speed up image analysis along with the improvement of the accuracy of results. About 98% overall accuracy and 0.001 quantization error in the recognition of small linear-trending bedforms demonstrate a promising framework.

Modeling Neutrosophic Data by Self-Organizing Feature Map: MANETs Data Case Study

Network security is a major research area for both scientists and business. Intrusion Detection System (IDS) is one of the most challenging problems in Mobile Ad Hoc Networks (MANETs). The main reason resides behind the changing and uncertain nature of MANETs networks. Hence, a compensate evolving in the IDS would be converting the whole system to rely on uncertainty and indeterminacy concepts. These concepts are the main issues in the fuzzy system and consequently in neutrosophic system. In neutrosophic system, each attack is determined by MEMEBERSHIP, INDETERMINACY and NONMEMEBERSHIP degrees. The main obstacle is that most data available are regular values which are not suitable for neutrosophic calculation. This paper is concerned by the preprocessing phase of the neutrosophic knowledge discovery system. Converting the regular data to neutrosophic values is a problem of generating the MEMEBERSHIP, NONMEMEBERSHIP and INDETERMINACY functions for each variable in the system. Self-Organized Feature Maps (SOFM) are unsupervised artificial neural networks that were used to build fuzzy MEMEBERSHIP function, hence they could be utilized to define the neutrosophic variable as well. SOFMs capabilities to cluster inputs using self-adoption techniques have been utilized in generating neutrosophic functions for the subsets of the variables. The SOFM are used to define the MEMEBERSHIP, NONMEMEBERSHIP and INDETERMINACY functions of the KDD network attacks data available in the UCI machine learning repository for further processing in knowledge discovery. Experimental results show the features and their corresponding functions.

Computational intelligence methodology for the analysis of RC circuit modelled with nonlinear differential order system

In this study, we solve nonlinear initial value problems arising in circuit analysis by applying bio-inspired computational intelligence technique using feed-forward artificial neural networks (ANNs) optimized with genetic algorithms (GAs), sequential quadratic programming (SQP), and their combined scheme. The system of resister–capacitor (RC) circuit having nonlinear capacitance is mathematically modelled with unsupervised ANNs by defining an energy function in mean-square error (MSE) sense. The objectives are to minimize the MSE for which the parameters of the networks are estimated initially with GA-based global search and in steady state with SQP algorithm for efficient local search. We consider a set of scenarios to evaluate the performance of the proposed scheme for different resistance and capacitance values along with current variations in the nonlinear RC circuit system. The results are compared with well-established fully explicit Runge–Kutta numerical solver in order to verify the accuracy of the applied bio-inspired heuristics. To prove the worth of the scheme, a comprehensive statistical analysis is provided for the performance metrics based on root MSE, mean absolute error, Theil’s inequality coefficient, Nash–Sutcliffe efficiency, variance account for, and the coefficient of determination (R 2).

Mapping energy-efficient technological advances in home appliances

The present study employs patent data on three groups of large energy-efficient appliances (i.e. freezers/refrigerators, washing machines and dishwashers) and provides a methodology for (i) mapping components related to energy efficiency improvements, (ii) mapping their evolution over time and (iii) testing the technological fungibility of these components. Our analysis model exploits an original patent selection process and the concept of technological relatedness using co-occurrence analysis of patent classes as input for self-organising maps (SOMs), an unsupervised artificial neural network able to represent high-dimensional data in a visually attractive and two-dimensional distance-based map. The results confirm the pervasive nature of energy efficiency to be nested in many technological components. In addition, we show that a dematerialisation process has affected the evolution of energy efficiency technologies over time, in a technology space characterised by a high level of complexity and variety. Energy efficiency has links with information technologies embodied in the appliances, which are far from their maturity path. Consequently, innovation and information policies are of utmost importance in order to achieve more ambitious energy efficiency targets in the groups of appliances analysed.

An AI based approach to multiple census data analysis for feature selection

The census of population and dwellings undertaken by national state institutions world over at regular time intervals, is a fantastic source of information. However, there are major challenges to overcome when transforming the census data that usually consists of a vast number of attributes, into useful knowledge. In this paper, an artificial intelligent (AI) based approach is investigated to select appropriate attribute features that indicate interesting patterns in Beppu census wards in 2000 and 2010. The results of the self-organising map or SOM (unsupervised artificial neural network) based clustering, GIS visualisation and machine learning (J48 and JRip functions of WEKA), provide relevant discerning features, new patterns and new knowledge that can be of use to many professionals, such as urban/transport planers and resources management.

Multiscale characterization of pore structure in carbonate formations: Application to the Scurry Area Canyon Reef Operators Committee Unit

Carbonate formations consist of a wide range of pore types with different shapes, pore-throat sizes, and varying levels of pore-network connectivity. Such heterogeneous pore-network properties affect the fluid flow in the formation. However, characterizing pore-network properties (e.g., effective porosity and permeability) in carbonate formations is challenging due to the heterogeneity at different scales and complex pore structure of carbonate rocks. We have developed an integrated technique for multiscale characterization of carbonate pore structure based on mercury injection capillary pressure (MICP) measurements, X-ray micro-computed tomography (micro-CT) 3D rock images, and well logs. We have determined pore types based on the pore-throat radius distributions obtained from MICP measurements. We developed a new method for improved assessment of effective porosity and permeability in the well-log domain using pore-scale numerical simulations of fluid flow and electric current flow in 3D micro-CT core images obtained in each pore type. Finally, we conducted petrophysical rock classification based on the depth-by-depth estimates of effective porosity, permeability, volumetric concentrations of minerals, and pore types using an unsupervised artificial neural network. We have successfully applied the proposed technique to three wells in the Scurry Area Canyon Reef Operators Committee (SACROC ) Unit. Our results find that electrical resistivity measurements can be used for reliable characterization of pore structure and assessment of effective porosity and permeability in carbonate formations. The estimates of permeability in the well-log domain were cross-validated using the available core measurements. We have observed a 34% improvement in relative errors in well-log-based estimates of permeability, as compared with the core-based porosity-permeability models.

Background Modeling to Detect Foreground Objects Based on ANN and Spatio-Temporal Analysis

<p>This paper presented an approach to building background model for moving object detection using unsupervised Artificial Neural Network (ANN) without any prior knowledge about foreground objects. First, using Local Binary Pattern (LBP) which is texture feature, builds a statistical Background Model using ANN, then, comparing the behavior of next incoming frame with model and decide each pixel whether is deviating from a model or not. And based on if method detects foreground objects then background model is updated to make this model adaptive. Also, spatial-temporal information has been exploited in this method to suppress sudden illumination variation and to suppress false foreground pixels. It was demonstrated and proved, by qualitative and quantitative metrics that the newly presented approach is adaptive, generic and can address all issues and challenges for background subtraction. To evaluate the performance of the presented approach this paper compared with recent approaches by using standard metrics and proved that presented method outperforms many existing recent approaches.</p>

An unsupervised artificial neural network versus a rule-based approach for fault detection and identification in an automated assembly machine

Artificial neural networks (ANNs) are suitable for fault detection and identification (FDI) applications because of their pattern recognition abilities. In this study, an unsupervised ANN based on Adaptive Resonance Theory (ART) is tested for FDI on an automated O-ring assembly machine testbed, and its performance and practicality are compared to a conventional rule-based method. Three greyscale sensors and two redundant limit switches are used as cost-effective sensors to monitor the machine's assembly process. Sensor data are collected while the machine is operated under normal condition, as well as 10 fault conditions. Features are selected from the raw sensor data, and data sets are created for training and testing the ANN. The performance of the ANN for detecting and identifying known, unknown and multiple faults is evaluated; the performance is compared to a conventional rule-based method using the same data sets. Results show that the ART ANN is able to achieve excellent fault detection performance with minimal modeling requirements; however, the performance depends on careful tuning of its vigilance parameter. Although the rule-based system requires more effort to set up, it is judged to be more useful when unknown or multiple faults are present. The ART network creates new outputs for unknown and multiple fault conditions, but it does not give any more information as to what the new fault is. By contrast, the rule-based method is able to generate symptoms that clearly identify the unknown and multiple fault conditions. Thus, the rule-based method is judged to be the most feasible method for FDI applications.

Tailwater habitat stability after dam removal and return to a natural hydrological regime

Abstract The study was conducted from 2000 to 2003 in the tailwater of the Drzewieckie Lake, an artificial reservoir in Central Poland. Short-term peaks in water flow were generated for the purpose of the operation of a whitewater slalom canoeing track built just downstream of the dam. In 2002, the reservoir was drawn down. The patterns in habitat samples were recognized with a Kohonen’s unsupervised artificial neural network (SOM). The SOM spatial gradient was stronger than the SOM temporal gradient, which shows that the removal of the studied dam did not have a destructive impact on habitats’ features, as shown in other studies, and that the patchy nature of the riverbed has been maintained. The complete emptying of the Drzewieckie Lake took place at the beginning of the vegetation season, which allowed plants to cover the exposed bottom of the reservoir and, consequently, reduce the downstream flow of organic matter accumulated there. Patterns in the displacement of aquatic macrophytes, inorganic substratum and different fractions of particulate organic matter are discussed. The amount of dissolved oxygen decreased because of the lack of intensive water discharge from the reservoir into the river, which would result in high water turbulence. Results of this study are important for planning the ecologically sound dam removals.

A New Method for Human Posture Recognition Using Principal Component Analysis and Artificial Neural Network

The recognition of human posture from images is currently a very active area of research in computer vision. This paper presents a novel recognition method to determine a human posture is of walking or sitting using Principal Component Analysis (PCA) and Artificial Neural Network (ANN). In this paper, two types of learning are used to recognize the human posture. One is unsupervised and another is supervised learning. We have used PCA for unsupervised learning and ANN for supervised learning. To evaluate the performance of the proposed method, we have considered four types of human posture; walking, sitting, right leg up-down and left leg up-down. The experimental results on the human action of walking, sitting, right leg up-down and left leg up-down database show that our approach produces accurate recognition.

Fen ecosystem responses to water‐level fluctuations during the early and middle Holocene in central Europe: a case study from Wilczków, Poland

This paper presents a palaeoenvironmental reconstruction of the Wilczków fen (central Poland). The fen developed in an inactive valley at the onset of the Holocene (~11 ka BP) and peat accumulation lasted until 5.7 ka BP. Multi‐proxy reconstructions were made on the basis of palaeobotanical, cladoceran, chironomid, beetle and geochemical analyses. A Kohonen self‐organizing map (SOM, unsupervised artificial neural network) of the biotic sequence distinguished four stages of fen history. Stage X1 (11.0–10.7 ka BP) was relatively wet and cool. Organic matter started to accumulate but the habitat conditions remained unstable. Moss, sedge and fern communities then developed. Sedimentary changes reveal an intensive groundwater supply at that time. Numerous and diverse chironomid and cladoceran subfossils indicate nearly permanent aquatic conditions. During stage Y1 (10.6–9.2 ka BP) conditions were dry and the upper peat layer desiccated. Cladocera nearly disappeared whereas chironomids were represented by semi‐terrestrial and predatory (Tanypodinae) species. Conditions started to be more reducing. All the remaining samples belonged to the interweaving stages X2 and Y2. Stage Y2 (mostly 9.1–7.3 and 6.0–5.7 ka BP) was also dry but humidity increased towards the top. Oxidizing conditions occurred and the pH became more alkaline, favouring Cladium mariscus. The basin received mostly allochthonous matter input at that time. Stage X2 (mostly 6.8–6.1 ka BP) was humid and warm. The groundwater supply remained low but there was an increase in precipitation, changing local conditions to ombrotrophic. Species‐rich chironomid and cladoceran communities were associated with temporary pools. Finally, conditions returned to those characteristic of stage Y2. The presented reconstruction documents long‐term abiotic and biotic changes determined by water supply, including groundwater outflow, which have rarely been detected at a multi‐proxy scale. We show that inactivated valley fens are sensitive to climate‐driven hydrological fluctuations. Kohonen neural networks appear to be a promising method for analysing variability in multi‐proxy data.

An improved I-FAST system for the diagnosis of Alzheimer's disease from unprocessed electroencephalograms by using robust invariant features.

This paper proposes a new, complex algorithm for the blind classification of the original electroencephalogram (EEG) tracing of each subject, without any preliminary pre-processing. The medical need in this field is to reach an early differential diagnosis between subjects affected by mild cognitive impairment (MCI), early Alzheimer's disease (AD) and the healthy elderly (CTR) using only the recording and the analysis of few minutes of their EEG. This study analyzed the EEGs of 272 subjects, recorded at Rome's Neurology Unit of the Policlinico Campus Bio-Medico. The EEG recordings were performed using 19 electrodes, in a 0.3-70Hz bandpass, positioned according to the International 10-20 System. Many powerful learning machines and algorithms have been proposed during the last 20 years to effectively resolve this complex problem, resulting in different and interesting outcomes. Among these algorithms, a new artificial adaptive system, named implicit function as squashing time (I-FAST), is able to diagnose, with high accuracy, a few minutes of the subject's EEG track; whether it manifests an AD, MCI or CTR condition. An updating of this system, carried out by adding a new algorithm, named multi scale ranked organizing maps (MS-ROM), to the I-FAST system, is presented, in order to classify with greater accuracy the unprocessed EEG's of AD, MCI and control subjects. The proposed system has been measured on three independent pattern recognition tasks from unprocessed EEG tracks of a sample of AD subjects, MCI subjects and CTR: (a) AD compared with CTR; (b) AD compared with MCI; (c) CTR compared with MCI. While the values of accuracy of the previous system in distinguishing between AD and MCI were around 92%, the new proposed system reaches values between 94% and 98%. Similarly, the overall accuracy with best artificial neural networks (ANNs) is 98.25% for the distinguishing between CTR and MCI. This new version of I-FAST makes different steps forward: (a) avoidance of pre-processing phase and filtering procedure of EEG data, being the algorithm able to directly process an unprocessed EEG; (b) noise elimination, through the use of a training variant with input selection and testing system, based on naïve Bayes classifier; (c) a more robust classification phase, showing the stability of results on nine well known learning machine algorithms; (d) extraction of spatial invariants of an EEG signal using, in addition to the unsupervised ANN, the principal component analysis and the multi scale entropy, together with the MS-ROM; a more accurate performance in this specific task.

Rock classification in the Haynesville Shale based on petrophysical and elastic properties estimated from well logs

Rock classification can enhance fracture treatment design for successful field developments in organic-shale reservoirs. The petrophysical and elastic properties of formations are important to consider when selecting the best candidate zones for fracture treatment. Rock classification techniques based on well logs can be advantageous compared to conventional ones based on cores, and they enable depth-by-depth formation characterization. We developed and evaluated three rock classification techniques in organic-shale formations that incorporate well logs and well-log-based estimates of elastic properties, petrophysical properties, mineralogy, and organic richness. The three rock classification techniques include (1) a 3D crossplot analysis of organic richness, volumetric concentrations of minerals, and rock brittleness index, (2) an unsupervised artificial neural network (ANN), built from an input of well logs, and (3) an unsupervised ANN, constructed using an input of well-log-based estimates of petrophysical, compositional, and elastic properties. A so-called self-consistent approximation rock-physics model is used to estimate elastic rock properties. This model enables assessment of the elastic properties based on the well-log-derived estimates of mineralogy and shapes of rock components, in the absence of acoustic-wave velocity logs. Finally, we apply the three proposed techniques to the Haynesville Shale for rock classification. We verify the identified rock types using thin-section images and previously identified lithofacies. We determined that well logs can be directly used for rock classification instead of petrophysical, compositional, and elastic properties obtained from well-log interpretation. Direct use of well logs, instead of well-log-derived properties, can reduce uncertainty associated with the physical models used to estimate elastic moduli and petrophysical/compositional properties. The three proposed well-log-based rock classification techniques can potentially enhance fracture treatment for production from complex organic-shale reservoirs through (1) detecting the best candidate zones for fracture treatment and (2) optimizing the number of required fracture stages.

AN INTELLIGENT HYBRID MULTI CRITERIA DECISION MAKING TECHNIQUE TO SOLVE A PLANT LAYOUT PROBLEM

Multi criteria decision making (MCDM) techniques in today's organizations, as a key to performance measurement comes more to the foreground with the advancement in the high technology. During recent years, many studies have been conducted to obtain a ranking among many alternatives via measuring performance of each of them against many criteria. Managerial decision making problems like supplier selection, weapon selection, project selection, site selection etc are dealt with many multi criteria decision making methods like TOPSIS, AHP-TOPSIS (Technique for Order Preference by Similarity to Ideal Solution), PROMETHEE (Preference Ranking Organization Method for Enrichment Evaluation), ELECTRE, VIKOR etc in crisp throughout the literature. In this work, we first compare several MCDM methodologies to validate the consistency of them on a standard dataset of plant layout problem. We proposed M-TOPSIS, A-TOPSIS procedure to select a suitable layout for the comparative study. Results of M-TOPSIS and A-TOPSIS have been employed to build an unsupervised artificial neural network (ANN) to obtain a new ranking of alternatives. This study proposes an approach of deriving the rank value, in order to get optimal configuration, from the average of more than one set of rank results obtained through the deployment of MCDM methodologies.