Levenberg-Marquardt Neural Network Research Articles

Predicting trait impressions of faces using local face recognition techniques

The aim of this work is to propose a method for detecting the social meanings that people perceive in facial morphology using local face recognition techniques. Developing a reliable method to model people’s trait impressions of faces has theoretical value in psychology and human–computer interaction. The first step in creating our system was to develop a solid ground truth. For this purpose, we collected a set of faces that exhibit strong human consensus within the bipolar extremes of the following six trait categories: intelligence, maturity, warmth, sociality, dominance, and trustworthiness. In the studies reported in this paper, we compare the performance of global face recognition techniques with local methods applying different classification systems. We find that the best performance is obtained using local techniques, where support vector machines or Levenberg-Marquardt neural networks are used as stand-alone classifiers. System performance in each trait dimension is compared using the area under the ROC curve. Our results show that not only are our proposed learning methods capable of predicting the social impressions elicited by facial morphology but they are also in some cases able to outperform individual human performances.

Fusion of systems for automated cell phenotype image classification

Automated cell phenotype image classification is related to the problem of determining locations of protein expression within living cells. Localization of proteins in cells is directly related to their functions and it is crucial for several applications ranging from early diagnosis of a disease to monitoring of therapeutic effectiveness of drugs. Recent advances in imaging instruments and biological reagents have allowed fluorescence microscopy to be extensively used as a tool to understand biology at the cellular level by means of the visualization of biological activity within cells. However, human classification of fluorescence cell micrographs is still subjective and very time consuming, thus an automated approach for the systematic determination of protein subcellular locations from fluorescence microscopy images is required. Existing approaches concentrated on designing a set of optimal features and then applying standard machine-learning algorithms. This paper takes into consideration the best methods proposed in the literature and focuses on the study of ensemble machine learning techniques for cell phenotype image classification. Two techniques are tested for the classification: a random subspace of Levenberg–Marquardt neural networks and a variant of the AdaBoost. Each of these two methods are tested with different feature sets, moreover the fusion between the two ensembles is studied. The best ensemble tested in this work obtains an outstanding 97.5% accuracy in the 2D-Hela dataset, which to the best of our knowledge is the best performance obtained on this dataset (the most used benchmark for comparing automated cell phenotype image classification approaches).

On-line security monitoring and analysis using Levenberg-Marquardt algorithm-based Neural Network

Due to open access in the restructured power system, the events of bus voltage limit violation and transmission line overloading are occurring frequently. These events are mainly responsible for several incidents of major network collapses leading to partial or even complete blackouts and due to this, security monitoring and analysis has become a challenging task to be performed at energy control centre. A fast and accurate method of Power Flow (PF) study may be able to investigate the system security by determining the power system static states, i.e. voltage magnitude and voltage angle at each bus. In this paper, a Levenberg-Marquardt algorithm-based Neural Network (LMNN) has been proposed which provides a fast learning to the multi-layer neural network. The effectiveness of the proposed LMNN-based approach for security monitoring and analysis has been demonstrated by computation of bus voltage magnitudes and voltage angles for line-outage contingencies at different loading conditions in IEEE 14-bus system.

Hybrid neuro-fuzzy system for power generation control with environmental constraints

The real time controls at the central energy management centre in a power system, continuously track the load changes and endeavor to match the total power demand with total generation in such a manner that the operating cost is least. However due to the strict government regulations on environmental protection, operation at minimum cost is no longer the only criterion for dispatching electrical power. The idea behind the environmentally constrained combined economic dispatch formulation is to estimate the optimal generation allocation to generating units in such a manner that fuel cost and harmful emission levels are both simultaneously minimized for a given load demand. Conventional optimization techniques are cumbersome for such complex optimization tasks and are not suitable for on-line use due to increased computational burden. This paper proposes a neuro-fuzzy power dispatch method where the uncertainty involved with power demand is modeled as a fuzzy variable. Then Levenberg–Marquardt neural network (LMNN) is used to evaluate the optimal generation schedules. This model trains almost hundred times faster that the popular BP neural network. The proposed method has been tested on two test systems and found to be suitable for on-line combined environmental economic dispatch.

APPLICATION OF REMOTE SENSING TO ESTIMATE ABOVE GROUND BIOMASS IN TROPICAL FORESTS OF INDONESIA

This work aims to estimate Above Ground biomass (AGB) of a tropical rainforest in East Kalimantan, Indonesia using equation derived from the stand volume prediction and to study the spatial distribution of AGB over aforest area. The potential of remote sensing and field measurement data to predict stand volume and AGB were studied Landsat ElM data were atmospherically corrected using Dark Object Subtraction (DOS) technique, and topographic corrections were conducted using C-correction method Stand volume was estimated using field data and remote sensing data using Levenberg-Marquardt neural networks. Stand volume data was converted into the above ground biomass using available volume - AGB equations. Spatial distribution of the AGB and the error estimate were then interpolated using kriging. Validated with observation data, the stand volume estimate showed integration of field measurement and remote sensing data has better prediction than the solitary uses of those data. The AGB estimate showed good correlations with stand volume, number of stems, and basal area.

유량과 수질을 연계한 실시간 인공지능 경보시스템 개발 (I) 유량-수질 예측모형의 적용

평창강 수질자동측정망 실시간 자료를 이용하여 강우시와 무강우시로 구분하여 분석하였다. 강우시에 측정된 TOC 자료는 무강우시 측정된 자료에 비해 평균값, 최대값, 표준편차가 크게 나타났으며, 강우시의 DO 자료는 무강우시에 측정된 자료보다 낮아 유량이 수질변화에 영향을 미치는 것으로 분석되었다. 신경망 모형과 뉴로-퍼지 모형으로 수질예측 모형을 구성하고, 적용하였다. LMNN, MDNN, ANFIS 모형은 TOC 모의에서 DO 예측에서는 LMNN, MDNN 모형이 ANFIS 모형보다 좋은 결과를 보였으며, 정량적 자료에 정성적 자료인 시간을 학습한 MDNN 모형이 가장 작은 오차를 보였다. 하천의 실시간적 관리를 위해서는 유량과 수질의 측정이 동일한 지점에서 동시간적으로 이루어져야 보다 효과적이다. 그러나 수질자동측정망 지점과 T/M 수위관측소가 원거리에 위치한 경우들이 있으며, 평창강 수질자동측정망 지점이 그 중 하나이다. 연구에서는 평창강 수질자동측정망 지점의 유출예측을 위한 신경망 모형을 구성하여 수질예측 모형과 연계하였으며, 연계된 모형은 수질예측에 개선된 결과를 보였다. It is used water quality data that was measured at Pyeongchanggang real time monitoring stations in Namhan river. These characteristics were analyzed with the water qualify of rainy and nonrainy periods. TOC (Total Organic Carbon) data of rainy periods has correlation with discharge and shows high values of mean, maximum, and standard deviation. DO (Dissolved Oxygen) value of rainy periods is lower than those of nonrainy periods. Input data of the water quality forecasting models that they were constructed by neural network and neuro-fuzzy was chosen as the reasonable data, and water qualify forecasting models were applied. LMNN, MDNN, and ANFIS models have achieved the highest overall accuracy of TOC data. LMNN (Levenberg-Marquardt Neural Network) and MDNN (MoDular Neural Network) model which are applied for DO forecasting shows better results than ANFIS (Adaptive Neuro-Fuzzy Inference System). MDNN model shows the lowest estimation error when using daily time, which is qualitative data trained with quantitative data. The observation of discharge and water quality are effective at same point as well as same time for real time management. But there are some of real time water quality monitoring stations far from the T/M water stage. Pyeongchanggang station is one of them. So discharge on Pyeongchanggang station was calculated by developed runoff neural network model, and the water quality forecasting model is linked to the runoff forecasting model. That linked model shows the improvement of waterquality forecasting.

Performance of the Levenberg–Marquardt neural network training method in electronic nose applications

The focus of this study is to find the appropriateness of the Levenberg–Marquardt (LM) neural network (NN) training algorithm for recognition of odor patterns associated with an electronic nose (e-nose). Multiple time-patterns represent step response of the array of sensors to the odorants. The experiments are performed on four representative classes of odorants: coffees, fragrances, hog farm air, and cola beverages. The odor recognition system is composed of a Karhunen–Loéve (KL) based pre-processing unit, and a feedforward neural network with the LM training algorithm. The parameters of the pre-processing unit and the neural network are fine-tuned using a genetic algorithm. Back-propagation algorithm with adaptive learning rate is selected as a standard neural network training method, for the purpose of comparison. The results of the experiments indicate that the LM algorithm provides high correct recognition ratios. In addition, the results confirm that the LM method outperforms the back-propagation (BP) method with adaptive learning rate, for the classes of the odorants provided in this study.

Molecular parameters responsible for the melting point of 1,2,3-diazaborine compounds.

Quantitative structure-property relationships (QSPR) have been determined to predict the melting point temperatures of 1,2,3-diazaborine compounds (n = 72). Electronic and topological descriptors were computed from molecular structures, and a QSPR model was generated by linear multiple regression using reported melting point temperatures as the dependent variable. The most important molecular descriptors describing this physicochemical property were the sum of the atomic charges for the heteroatoms, the sum of Randić connectivity indexes (0)Chi(0), (0)Chi(1), and (0)Chi(2), the total number of atoms in the molecule, and the volume of the box in which the molecule fits. The multiple determination coefficient (R(2)) and standard error of estimation for the model were 0.856 and 16.787 degrees C, respectively. In addition to regression techniques, a back-propagation neural network was used to include nonlinear relationships between molecular structure and melting point temperatures. It is concluded that melting point temperatures for 1,2,3-diazaborine compounds can be described by electrostatic interactions mediated by atomic charges and steric properties. The results of this study demonstrate that multiple linear regression analysis and back-propagation neural network are techniques that can be used to successfully predict the melting point temperatures of 1,2,3-diazaborine compounds. The most accurate prediction results were obtained using the Levenberg-Marquardt neural network algorithm.