Search

This paper analyzes the possibility of applying data fusion combined with artificial neural networks (ANN) on a dataset combining hard and soft data for prediction of one of the most devastating crop diseases of winter wheat, i.e., Septoria Tritici (Zymoseptoria tritici). In advanced decision support systems for crop protection choices, disease models form a major component.They reproduce the biophysical processes of disease development and temporal spread as a set of rules or processes to predict disease risk value. However, the adaptation of these rules or processes to incorporate the effects of climate change is complex and requires extensive rework. To remedy this issue, statistical machine learning techniques have been introduced to model disease severity percentage for some diseases.However, the use of artificial neural networks has been limited (mainly to image data) and is unexplored for Septoria Tritici.This paper explores the use of Feed Forward neural networks on fused tabular data for the task of disease severity modelling. First, ten years of trial data ranging from 2008 to 2018 across Europe is used for the creation of the new tabular dataset with a fusion of all important data sources baring impact on disease development: Field-specific data, weather data, crop growth stages, and disease severity observation made by human trial operators (response variable). %Correlation and regression analyses and domain expert knowledge were used for the selection of useful predictor variables from these data sources. Next, two implementation architectures of Feed Forward neural networks on tabular data are employed: a) standard architecture with backpropagation, drop out regularization, and batch normalization and b) advanced architecture with improvements such as cyclic learning rate and cosine annealing.%A comparison of generic two layer feed forward network vs the same with the incorporation of the latest advances to improve the performance of the architecture is presented. The advanced architecture is able to better model the data and make estimations of disease severity with a difference of +-10\% giving a better quantifiable estimate of disease stress. For better outreach to farmers, a technique to incorporate such modelling techniques into the well established Decision Support Systems is also presented. ISBN för värdpublikation: 978-0-578-64709-8, 978-1-7281-6830-2

This paper proposes an Artificial Neural Network (ANN) to accurately predict the real angles obtained by a Triangulation Vision System. The performance of the ANN is compared with the K-Nearest Neighbors algorithm from previous publications. For the experimentation it was necessary to create a database to train and prove both methods in different coordinates on a determinate area through the dynamic triangulation method. Afterwards, the root mean square error is calculated to obtain the accuracy of each algorithm. Finally, several laser scanning measurements were taken at different distances to analyze the measurement dispersion of both algorithms. This paper proposes an Artificial Neural Network (ANN) to accurately predict the real angles obtained by a Triangulation Vision System. The performance of the ANN is compared with the K-Nearest Neighbors algorithm from previous publications. For the experimentation it was necessary to create a database to train and prove both methods in different coordinates on a determinate area through the dynamic triangulation method. Afterwards, the root mean square error is calculated to obtain the accuracy of each algorithm. Finally, several laser scanning measurements were taken at different distances to analyze the measurement dispersion of both algorithms.

The short-term forecasting of electricity consumption has emerged as an important area of research with the aim of increasing the efficiency and reliability of energy system operation. It plays a very important role in the field of scheduling, load analysis, planning and maintenance of the power system. Artificial Neural Networks (ANNs) are computational techniques which have been successfully applied in a number of different problems, including short-term electricity consumption forecasting. This paper gives an overview of the recently published research papers that deal with short-term electricity consumption forecast based on ANNs. The related papers are evaluated with respect to several aspects, such as applied ANN type, used input variables and obtained precision regarding short-term electricity consumption prediction.

Predicting energy demand peak is a key factor for reducing energy demand and electricity bills for commercial customers. Features influencing energy demand are many and complex, such as occupant behaviours and temperature. Feature selection can decrease prediction model complexity without sacrificing performance. In this paper, features were selected based on their multiple linear regression correlation coefficients. This paper discusses the capabilities of M5 model trees in energy demand prediction for commercial buildings. M5 model trees are similar to regression trees, however they are more suitable for continuous prediction problems. The M5 model tree prediction was developed based on a selected feature set including sensor energy demand readings, day of the week, season, humidity, and weather conditions (sunny, rain, etc.). The performance of the M5 model tree was evaluated by comparing it to the support vector regression (SVR) and artificial neural networks (ANN) models. The M5 model tree outperformed the SVR and ANN models with a mean absolute error (MAE) of 8.94 compared to 10.02 and 12.04 for the SVR and ANN models respectively.

Level-crossing ADCs reduce the size of data streams in wearable devices. However, in the context of electrocardiogram (ECG) signals, such an event-driven data source results in a variable length two-dimensional (time-amplitude tuples) data vector for each ECG beat. It is difficult to apply many standard signal processing techniques to this data making classifiers more complex. In this paper we resolve these difficulties by mapping the variable length 2D vectors to a fixed length feature vector comprising the first 81 coefficients of a Chebyshev polynomial expansion of the ECG beat. We show that beat reconstruction based on these 81 coefficients results in an average RMS difference to the original beat of only ≈ 3.08%. Using these coefficients as the feature set input to a simple three-layered ANN binary (Normal / Abnormal) ECG classifier and we demonstrate 98.15% average accuracy and 96.07% average sensitivity. Using the same simple ANN structure we also constructed a 4-class ANN structure which achieved 98.80% average accuracy and 91.5% average sensitivity. Both these networks have only 20k parameters and outperform the state-of-the-art classifiers, enabling low-power edge computing.

En este artículo se presenta una metodología para estimar la curva de carga de transformadores de potencia utilizando redes neuronales artificiales. Para implementar la metodología propuesta se utilizaron datos reales de dos transformadores ubicados en diferentes zonas geográficas y con diferentes condiciones de operación. La técnica de estimación de carga fue implementada con datos históricos de la empresa Interconexión Eléctrica S.A (ISA). Para realizar la predicción de las curvas de carga se entrenó una red neuronal usando MATLAB, la cual se adapta a la curva de carga con tiempos de predicción diaria y semanal. La metodología propuesta permite la estimación de la curva de carga de transformadores de potencia con un error relativo de alrededor del 10%. Este método puede ser utilizado en otros equipos con características de operación similares. This paper presents a methodology for load estimation of power transformers by means of arfificial neural networks. To implement the proposed methodology the data of two power transformers, located in different geographical areas and wifh different operational conditions, were considered. The load estimation methodology was implemented using historical data provided by the company Interconexión Eléctrica S.A. (ISA). To predict the load curves a neural network was trained using MATLAB, being able to fit a load curve with daily and weekly prediction times. The proposed method allows the estimation of load curve values in power transformers with an average percent relative error of around 10%. The method described in this paper can be applied fo other equipment with similar operating characteristics.

International audience; Interested in Evolutionary Robotics, this paper focuses on the acquisition and exploitation of memory skills. The targeted task is a well-studied benchmark problem, the Tolman maze, requiring in principle the robotic controller to feature some (limited) counting abilities. An elaborate experimental setting is used to enforce the controller generality and prevent opportunistic evolution from mimicking deliberative skills through smart reactive heuristics. The paper compares the prominent NEAT approach, achieving the non-parametric optimization of Neural Nets, with the evolutionary optimization of Echo State Networks, pertaining to the recent field of Reservoir Computing. While both search spaces offer a sufficient expressivity and enable the modelling of complex dynamic systems, the latter one is amenable to robust parametric, linear optimization with Covariance Matrix Adaptation-Evolution Strategies.

In this paper, two different artificial neural networks are tested and compared with regard to their application in the estimation of difficult- to-measure process variables. Two of the most commonly used neural networks, the MLP (multi- layer perceptron) and RBF (radial basis function) neural networks, with simple structure and standard training methods are chosen as examples. Neural network training is based on available data from a database of process variables measured over a long time period. The database in this paper is obtained using a simulation model of a real process. Without going deeper into theoretical background, relative properties of these neural networks are given through the results obtained by testing the trained networks and analysis performed on these results.

The paper presents recent advances in the design of controllable, highly accurate, and multi–band Raman gain profiles. The ultra–wideband programmable gain profiles are implemented using a machine learning approach based on the mapping between gain profiles and pump powers.

Although many mathematical and analytical techniques have been presented to control and identify the dynamic systems, there are vast fields of research needing to be developed and extended through Deep Learning (DL) approaches. In this paper, we try to describe how intelligent controllers can interact under control systems in a unique DL-based package. Despite the fact that conventional techniques have some advantages, such as the appropriate reliability and simple implementation for industrial goals, intelligent methods have potential to solve complex problems and identify nonlinear systems. Hence the concentration of this research is on the use of DL techniques to improve the system identification and control in model reference adaptive controllers. A dataset is also used to validate the responses of the proposed techniques. The simulation results demonstrate that not only are the proposed methods consistently appropriate to control the complex systems but also they have acceptable responses in order to utilize for system identification.

Touch-screens have become the most common way of interaction between humans and machines. Nevertheless, the lack of rich haptic-feedback on most of the devices limits the quality and effectivity of the interaction. To develop novel haptic feedback generation strategies, the authors of this work, are initially experimenting with machine learning algorithms to detect the position of a finger over a tactile surface. This paper presents the experimental study to determine the effect of the impact contact duration on the different machine learning (ML) and neural network (NN) models that were previously proposed for impact position detection. A new version of a Linear Impact Generator (LIG) is presented and an experimental study, with a high-speed camera, is carried out to characterize the LIG. Additionally, two different pre-processing methods are compared (i.e. Magnitude Spectrogram representation and FFT frequency-domain representation), showing that the FFT representation contains richer information to describe the impact position. The best model achieved an error (Validation MAE) of 0.18 % or 0.31 mm and (Test MAE) of 1.16 % or 2 mm. Lastly, it was demonstrated that the impact contact duration has a direct effect on the precision of the impact position prediction. When the contact duration changes, the error increases to (Test MAE) 27.12 % or 48 mm on average.

Breast cancer is one of the commonest cause of cancer deaths in women. It starts developing when threatening bumps start forming from the breast cells, and unfortunately most diagnoses happen in later stages, thus resulting in low chances of survival for the patient. So for early detection and prognosis, it is necessary to detect the benign or threatening nature of the bumps. In this paper, Artificial Neural Networks (ANN) and Decision Tree (DT) classifiers are used to develop a machine learning (ML) model using the Wisconsin diagnostic breast cancer (WDBC) dataset, in order to evaluate the attributes of a breast cancer development at beginning phases and classify it as malignant or benign. In the proposed scheme, feature selection and feature extraction are done to extract statistical features from the dataset and comparison between the models is provided based on their performance to identify the most suitable approach for diagnosis. The dataset apportioned into various arrangements of train-test split. The presentation of the framework is estimated, depending on accuracy, sensitivity, specificity, precision, and recall. The binary classification problem achieved a maximum accuracy of 98.55%.

In recent times, thanks to the availability of a large quantity of data coming from the industrial process, several techniques based on a data-driven approach could be developed. Between all the data-driven techniques, as Principle Component Regression, Support Vector Machines, Artificial Neural Networks, Neuro-Fuzzy Systems, and many others, the data on which they rely should be analyzed to find correlations and dependencies that could improve their design. For this reason, the Input variable Selection (IVS) process has become of great interest in the recent period. The classical IVS relies on classical statistics, as Pearson coefficients, able to discover linear dependencies among data; today, due to the significant amount of data available, the challenge of also discovering non-linear dependencies appears to be a necessary skill, mainly for the design and development of a neural network. This paper proposes the use of a novel statistical tool named Maximal Information Coefficient (MIC) for developing an IVS procedure able to discover dependencies in a considerable dataset and guide the IVS designer to the selection of input variables in a data-driven application. As a case study, the procedure will be applied to a real application developed in the context of the Swedish forest industry, in order to choose the input variables of a neural network able to estimate the timber bundles volume, which represents an expensive parameter to measure in this context.

Vehicular accidents cause tragic loss of lives and traffic congestion to the transportation system. Therefore, prompt detection of traffic incidents offers tremendous benefits of minimizing congestion and reducing secondary accidents. Most incident management systems use inductive loop detectors for incident detection. Inductive loops are the most commonly used traffic detectors and they collect data such as vehicle speed at a point. However, the implemented algorithms using loop detectors showed mixed success. I think that the changes in average traffic speed in case of traffic incidents have certain patterns that are different from the normal conditions. In this paper, I try to automatically detect traffic incidents using artificial neural networks and traffic condition information of the traffic information center. In the field tests, the new model performed better than existing methods.

International audience; Inspired from the brain, neuromorphic computing would be the right alternative to traditional Von-Neumann architecture computing that knows its end of growth as predicted by Moore’s law. In this paper, we explore bio-inspired neural networks as an AI-accelerator for embedded systems. To do so, we first map neural networks from formal to spiking domain, then choose the information coding method resulting in better performances. Afterwards, we present the design of two different hardware architectures: time-multiplexed and fully-parallel. Finally, we compare their performances and their hardware costto select at the end the adequate architecture and conclude about spike-based neural networks as a potential solution for embedded artificial intelligence applications.

In many industrial fields, the direct measurement of a specific parameter of interest can be too costly to be feasible. Using data from other parameters about the process, a data driven approach can instead be used for a cost-efficient estimation of the parameter of interest. For the Swedish forest industry, the measurement of the timber bundles volume sets the price for the customers. Volume is in this case an expensive parameter to measure, where manual measurements have been used as reference to a remote imaging estimation system involving operators. In this paper a data driven approach is presented for the estimation of timber bundles volume, using correlated features such as weight, date of cut, storage time etc., to demonstrate a costefficient software estimation system. The proposed approach, based on neural networks, shows comparable results to the existing remote estimation method when using a few features and even better performance when adding feature information from the timber harvesters.

This paper presents a method that can be used for the efficient detection of small maritime objects. The proposed method employs aerial images in the visible spectrum as inputs to train a categorical convolutional neural network for the classification of ships. A subset of those filters that make the greatest contribution to the classification of the target class is selected from the inner layers of the CNN. The gradients with respect to the input image are then calculated on these filters, which are subsequently normalized and combined. Thresholding and a morphological operation are then applied in order to eventually obtain the localization. One of the advantages of the proposed approach with regard to previous object detection methods is that it is only required to label a few images with bounding boxes of the targets to be trained for localization. The method was evaluated with an extended version of the MASATI (MAritime SATellite Imagery) dataset. This new dataset has more than 7 000 images, 4 157 of which contain ships. Using only 14 training images, the proposed approach achieves better results for small targets than other well-known object detection methods, which also require many more training images. This work was supported in part by the Spanish Government's Ministry of Economy, Industry, and Competitiveness under Project RTC-2014-1863-8, and in part by the Babcock MCS Spain under Project INAER4-14Y (IDI-20141234).

The Publisher's final version can be found by following the DOI link. Distribution system is the crucial part of the power system which needs to be monitored for quality since it is the last stage in delivering electricity to consumers. The main aim of suppliers is therefore to maintain excellent quality of supply with minimal support at the network level. This includes dealing with issues around voltage regulation and minimization of real power losses on the network which are reflected in a uniform voltage profile through various load points on the feeders. However, with increase in decentralization, Distributed Generation (DG) becomes the focal point of distribution networks and can be used to achieve the above mentioned objectives. DGs are used in this paper to minimize the losses along with an improvement in the voltage profile. The location of these DG units on the network is addressed as an optimization problem using the loss sensitivity index and voltage deviation index. The size of power injected by DG units is obtained with using an intelligent approach that combines Genetic Algorithm (GA) with neural networks. Different cases of DG in terms of real and reactive power injected are considered resulting in an improvement in voltage profile. The loss sensitivity and voltage deviation indices show the best result for the location of DG needed to minimize the power loss along with improvement in voltage regulation. This approach is tested on a 13-bus distribution network with different types of loads placed throughout the feeder.

In the last few years, the reduction of energy consumption and pollution became mandatory. It became also a common goal of many countries. Only in Europe, the building sector is responsible for the total 40% of energy consumption and 36% of $CO_{2}$ pollution. Therefore, new control policies based on the forecast of buildings energy behaviors can be developed to reduce energy waste (i.e. policies for Demand Response and Demand Side Management). This paper discusses an innovative methodology for smart building indoor air-temperature forecasting. This methodology is based on a Non-linear Autoregressive neural network. This neural network has been trained and validated with a dataset consisting of six years indoor air-temperature values of a building demonstrator. In detail, we have studied three characterizing rooms and the whole building. Experimental results of energy prediction are presented and discussed.

Artificial Neural Networks (ANNs) constitute a research area of high interest, for both practitioners and academics, as they are found very useful for solving complex problems that are difficult to solve using known and well developed conventional methods or techniques. They can be used for prediction, control, estimation, data clustering and many other applications that are found in everyday scenarios. This paper explains in brief the basic theory of ANNs, followed by a review of different studies related to ANNs used for applications in buildings such as energy management, systems control and energy prediction. It has been found that applying ANNs in buildings the energy consumption can be reduced, depending on the application. Furthermore, efficient control mechanisms also become possible, leading to the reduction of the energy consumption. Through this review, the reader will be able to recognise the value of ANNs and their big potential in buildings and energy sector, in general. Finally, an ANN-based structure for predicting the local RES generation and the load demand for a building is proposed.

This paper describes a work in progress on co-evolving Artificial Neural Networks (ANNs) for music improvisation. Using this neuro-evolutionary approach the ANNs adapt to the changes in the human player's music as input, while still maintaining some of the structure of the musical piece previously evolved. The system is called PRIMAL-IMPROV and evolves modules that are composed of two ANNs, one controlling pitch and one controlling rhythm. The results of a quantitative study show that, by only introducing simple rules as fitness functions, the system is able to generate more interesting arrangements than ANNs evolved without a specific objective. The emerging and interesting musical patterns that are produced by the evolved ANNs hint at the promising potential of the system.

The application of Artificial Neural Networks (ANNs) to discriminate tag actions in UHF-RFID gate is presented in this paper. By exploiting Received Signal Strength Indicator values acquired in a real experimental scenario, a multi-layer perceptron neural network is trained to distinguish among tags incoming, outgoing or passing the RFID gate. A 99% accuracy can be obtained in tag classification by employing only one reader antenna and independently from tag orientation and typology.

In this paper we evaluate three state-of-the-art neural-network-based approaches for large-scale video classification, where the computational efficiency of the inference step is of particular importance due to the ever increasing amount of data throughput for video streams. Our evaluation focuses on finding good efficiency vs. accuracy tradeoffs by evaluating different network configurations and parameterizations. In particular, we investigate the use of different temporal subsampling strategies, and show that they can be used to effectively trade computational workload against classification accuracy. Using a subset of the YouTube-8M dataset, we demonstrate that workload reductions in the order of 10×, 50× and 100× can be achieved with accuracy reductions of only 1.3%, 6.2% and 10.8%, respectively. Our results show that temporal subsampling is a simple and generic approach that behaves consistently over the considered classification pipelines and which does not require retraining of the underlying networks., 2017 25th European Signal Processing Conference (EUSIPCO), ISBN:978-0-9928-6267-1

Various modalities are used for the examination of the gastrointestinal (GI) tract. One such modality is Wireless Capsule Endoscopy (WCE), a non-invasive technique which consists of a swallowable color camera that enables the detection of GI pathology with only minimal patient discomfort. Currently, tracking of the capsule position is estimated in the 3D abdominal space, using radio-frequency (RF) triangulation. The RF triangulation technique, however, does not provide sufficient information about the location of the capsule along the GI lumen, and consequently, the localization of any possible abnormality. Recently, we proposed a geometric visual odometry (VO) method for the localization of the capsule in the GI lumen. In this paper, we extend this state-of-art method by exploiting an artificial neural network (ANN) to augment the geometric method and achieve higher localization accuracy. The results of this novel approach are validated with an in-vitro experiment that provides ground truth information about the location of the capsule. The mean absolute error obtained, for a distance of 19.6cm, is 0.79±0.51cm.

Semi-active suspension control needs to measure the relative velocity of the wheels respect to the vehicle body to regulate the damping forces. Linear potentiometers are the most used sensors in racing for linearity and simplicity, but they are expensive for mass market and unreliable in the long run. The paper deals with two different approaches to design a soft sensor for the on-line estimation of the suspension velocity from acceleration signals. Main features of the soft sensor implementation according to each methodology as well as the corresponding experimental results are discussed with reference to an extensive measurement campaign carried out with a motorcycle test bed.

This paper presents a design of an artificial neural network (ANN) and feature extraction methods to identify two types of arrhythmias in datasets obtained through electrocardiography (ECG) signals, namely arrhythmia dataset (AD) and supraventricular arrhythmia dataset (SAD). No special ANN toolkit was used; instead, each neuron and necessary calculus were modeled and individually programmed. Thus, four temporal-based features are used: heart rate (HR), R-peaks root mean square (R-RMS), RR-peaks variance (RR-VAR), and QSR-complex standard deviation (QSR-SD). The network architecture presents four neurons in the input layer, eight in hidden layer and an output layer with two neurons. The proposed classification method uses the MIT-BIH Dataset (Massachusetts Institute of Technology-Beth Israel Hospital) for training, validation and execution or test phases. Preliminary results show the high efficiency of the proposed ANN design and its classification method, reaching accuracies between 98.76% and 98.91%, when in the identification of NSRD and arrhythmic ECG; and accuracies of 86.37% (AD) and 76.35% (SAD), when analyzing only classifications between both arrhythmias. info:eu-repo/semantics/acceptedVersion

This paper presents a method of creating a neurocontroller based on a multilayer perceptron for an unmanned aerial vehicle. We show how a neural network can effectively emulate dynamic characteristics of an aerial craft. Another network learns to control the emulator, using backpropagation algorithm to calculate the error in its control signal. A set of parameters is used to analyze the efficiency of the stabilization and the weights of the neurocontroller are adjusted accordingly. It is shown that the system meets stabilization requirements with sufficient number of iterations. Described method can be used to remotely control unmanned aerial vehicles operating in changing environment.

The aim of this paper is to develop a theoretical framework for training neural network (NN) models, when data is distributed over a set of agents that are connected to each other through a sparse network topology. The framework builds on a distributed convexification technique, while leveraging dynamic consensus to propagate the information over the network. It can be customized to work with different loss and regularization functions, typically used when training NN models, while guaranteeing provable convergence to a stationary solution under mild assumptions. Interestingly, it naturally leads to distributed architectures where agents solve local optimization problems exploiting parallel multi-core processors. Numerical results corroborate our theoretical findings, and assess the performance for parallel and distributed training of neural networks.

In this paper, detection method and classification technique of power quality disturbances is presented. Due to the increase of nonlinear load recently, it becomes an essential requirement to insure high level of power supply and efficient commotional consuming. Wavelet Transform represents a powerful mathematical platform which is needed especially at non-stationary situations. Disturbances are fed into wavelets to filter, detect and extract its features at different frequencies. Training of features extracted by DWT is done using artificial neural networks ANN to classify power quality disturbances.

Cardiopulmonary exercise testing is a non-invasive method widely used to monitor various physiological signals, describing the cardiac and respiratory response of the patient to increasing workload. Since this method is physically very demanding, innovative data analysis techniques are needed to predict patient response thus lowering body stress and avoiding cardiopulmonary overload. This paper proposes the Cardiopulmonary Response Prediction (CRP) framework for early predicting the physiological signal values that can be reached during an incremental exercise test. The learning phase creates different models tailored to specific conditions (i.e., single-test and multiple-test models). Each model can be exploited in the real-time stream prediction phase to periodically predict, during the test execution, signal values achievable by the patient. Experimental results on a real dataset showed that CRP prediction is performed with a limited and acceptable error.

2015 23rd Signal Processing and Communications Applications Conference, SIU 2015 -- 16 May 2015 through 19 May 2015 -- -- 113052 In this paper, an estimation of angle of hand opening-closing movements by using the Artificial Neural Network (ANN) from surface electromyography (sEMG) signal is presented. The first step of this method is to record sEMG signal from the subject's right forearm and to acquired video frames of hand at the same time. The second step is to synchronize the beginning and the end of recorded video frame and obtain sEMG signals. The third step is to extract some most commonly used feature vectors for sEMG in the literature. Finally, feature vectors sets are fed to the ANN to estimate angle of hand movements. The obtained success rate of the ANN is given as 94.06% in the train set and 93.41% in the test set. © 2015 IEEE.

Edge detection is a standard operation in image processing. It becomes problematic if noise is not additive, not Gaussian and not i.i.d. as this happens in images acquired by synthetic aperture radar (SAR). To perform edge detection better, it has been recently proposed to apply a trained neural network (NN) and SAR image pre-filtering for single-look mode. In this paper, we demonstrate that the proposed detector is, after certain modifications, applicable for edge detection in two-look and dual-polarization SAR images with and without pre-filtering. Moreover, we show that a recently introduced parameter AUC (Area Under the Curve) can be helpful in optimization of parameters for elementary edge detectors used as inputs of the NN edge detector. Quantitative analysis results confirming efficiency of the proposed detector are presented. Its performance is also studied for real-life TerraSAR-X data.

Association for Machine Learning and Applications (AML and A);IEEE Computer Society 2013 12th International Conference on Machine Learning and Applications, ICMLA 2013 -- 4 December 2013 through 7 December 2013 -- Miami, FL -- 104787 In this paper, a software system, which employs an intelligent approach to adjust learning process more accurately by determining student's learning status, is described briefly. The system comes with an Artificial Neural Networks based infrastructure to evaluate students' learning styles or levels before feeding its interface with the related course contents. The Artificial Neural Networks structure is mainly fed with answers that were given for a specially designed Multiple Intelligences test and this data is also combined with some other ones like examination grades, or points given by the course teacher for each student. Eventually, the stored course contents are then viewed to the active student, according to his / her learning status determined by the system. The designed and developed software system has been tested during the 'visual programming' course and obtained results have been also reported in this study. © 2013 IEEE.

International audience; This paper deals with an industrial application of a robotic assembly lines problem (RALB). It consists of seizing products on a moving conveyor and placing them on different deposit points. The goal is to optimize the efficiency of the line by assigning the suitable tasks and components to each robot. The performances evaluations of the system are done using a discret event simulation model. As in our industrial application we are bounded by the execution time, we suggest an adaptive harmony search (HS) to solve the above problem. Then, we propose to couple the algorithm with a guided local search (GLS), a fuzzy logic controller (FLC) and an artificial neural network (ANN) in order to enhance the efficiency of this method. All these methods allow avoiding local optimum solutions by improving the search ability and the parameters of the adaptive harmony search. After that, an exact method based on full enumeration is also developed to assess the quality of the developed methods. The experimental results show the advantages and the efficiency of the HS-GLS within a short computational time.

Combining several ANNs into ensembles normally results in a very accurate and robust predictive models. Many ANN ensemble techniques are, however, quite complicated and often explicitly optimize some diversity metric. Unfortunately, the lack of solid validation of the explicit algorithms, at least for classification, makes the use of diversity measures as part of an optimization function questionable. The merits of implicit methods, most notably bagging, are on the other hand experimentally established and well-known. This paper evaluates a number of straightforward techniques for introducing implicit diversity in ANN ensembles, including a novel technique producing diversity by using ANNs with different and slightly randomized link structures. The experimental results, comparing altogether 54 setups and two different ensemble sizes on 30 UCI data sets, show that all methods succeeded in producing implicit diversity, but that the effect on ensemble accuracy varied. Still, most setups evaluated did result in more accurate ensembles, compared to the baseline setup, especially for the larger ensemble size. As a matter of fact, several setups even obtained significantly higher ensemble accuracy than bagging. The analysis also identified that diversity was, relatively speaking, more important for the larger ensembles. Looking specifically at the methods used to increase the implicit diversity, setups using the technique that utilizes the randomized link structures generally produced the most accurate ensembles.

There are many proposals for binaural processing models and their applications for a hearing assistance system. These models attracts attention because of their directional selectivity in frontal direction of user. Not only binaural models but also any of two-element array system have a well know ambiguity in front-back discrimination which is called as “front-back confusion” or called as “cone of confusion” in psychoacoustics. Recently it was found that spectral cue of sound provides keys to solve this confusion in binaural listening condition and it was also reported the peaks and notches of spectral components play main role to estimate the vertical angle in sagittal coordinate. In this paper, a new method to estimate sound source direction on sagittal coordinate is proposed in order to solve the front-back confusion. It is implemented on an artificial neural network using interaural level and phase differences as input. The results of simulation using a set of Head Related Transfer Function of dummy head show that the averaged estimation error in quadrant segmentation is less than 1.0% for various types of sound.

International audience; Modulation type is one of the most important characteristics used in signal waveform identification and classification. In this paper, an algorithm for blind digital modulation identification for multiple-input multiple-output (MIMO) systems is proposed. The suggested algorithm is verified using higher order statistical moments and cumulants of the received signal. A multi-layer neural network trained with resilient backpropagation learning algorithm is proposed as a classifier. The purpose is to discriminate among different M-ary shift keying linear modulation types and the modulation order without any priori signal information. This study covers different MIMO systems with and without channel state information (CSI). The proposed classifier is evaluated through the probability of identification where we show that our proposed algorithm is capable of identifying the modulation scheme with high accuracy in excellent signal-to-noise ratio (SNR) range.

International audience; This paper analyzes a method for the DOA estimation of correlated signals based on the integration of a SVM classifier and an iterative multi-scaling approach (IMSA). The aim of the SVM classifier is the definition of a probability map of the angular directions of incidence of the incoming signals starting from the knowledge of the correlation matrix of the antenna array, while the IMSA procedure is used to reduce the computational load as well as to increase the angular resolution accuracy.

International audience; This paper presents an application of Artificial Neural Networks (ANNs) in the renewable energy domain and, more particularly, to predict solar energy. We look at the Multi-Layer Perceptron (MLP) network which has been the most used of ANNs architectures both in the renewable energy domain and in the time series forecasting. In previous studies, we have demonstrated that an optimized ANN with endogenous inputs can forecast the solar radiation on a horizontal surface with acceptable errors. Thus we propose to study the contribution of exogenous meteorological data to our optimized MLP and compare with different forecasting methods used previously: a naïve forecaster like persistence and an ANN with preprocessing using only endogenous inputs. Although intuitively the use of meteorological data may increase the quality of prediction, the obtained results are relatively mixed. The use of exogenous data generates a decrease of nRMSE between 0.5% and 1% for the two studied locations. The absolute error (RMSE) is decreased by 52 Wh/m²/day in the simple endogenous case and 335 Wh/m²/day for the persistence forecast.

International audience; The coupled computational intelligence, in term of genetic algorithms and artificial neural networks, with the time-domain solver for design of the microwave devices is presented in this paper, where the coupled genetic algorithms with the time-domain solver is used for optimization and the artificial neural networks is used to surrogate the time-domain solver in design of the microwave components. This microstrip patch antenna and the microstrip meander line filter are used for evaluation of the performance of computational intelligence in design microwave devices.

The paper proposes an estimator of the relevance of each weight connection on generalization which relies on a probabilistic description of the training phase: based on the above estimator, a pruning technique is derived.

In this paper, we present the application of artificial neural networks to the control of a mobile, autonomous robot, which is acting in a totally unknown and - most importantly - dynamically changing environment. In particular, the employment of interacting 'simple', i.e. hand-designed, neural networks for navigation purposes is investigated as well as a variation of self-organizing maps for adaptive sensor calibration. We insofar take a pragmatic point of view as the minimal condition imposed on the developed algorithms is that they do well on a real system acting in a real environment. Hence, the design of all of the implemented neural networks is clearly motivated by their applicability. In this context, special considerations are dedicated to ensure robustness, real-time capability and memory resourcefulness. In order to practically demonstrate the obtained results, the minirobot Khepera is utilized as an experimentatory platform, which is - due to its small size - a versatile tool for scientific investigation.

The objective of this paper is to provide an overview of the recent developments in the use of artificial neural networks in medical imaging. The areas of medical imaging that are covered include: ultrasound, magnetic resonance, nuclear medicine and radiological (including computerized tomography). 431 440 Sponsors: IEEE ARPA Conference code: 21520 Cited By :6