Search

In this paper an explanation of the structure and how a self-organized neuro-fuzzy inference system (SONFIS) works, is given with detail. The study uses three classification problems (Fisher iris, Breast Cancer and Human Activities) to then compare the results with well-known universal classifiers such as artificial neural networks (ANN) and multiclass support vector machines (SVM). A brief description of each of these methods is presented. The results show that SONFIS has a similar, and sometimes better, performance than ANN and SVM with the advantage of generating a rule basis that helps understanding the inner structure of the problem. [ABSTRACT FROM AUTHOR]

This paper presents the design and testing of a new training model for single hidden layer feedforward network based on the same properties of Extreme Learning Machine (ELM). The model acts by compressing the information coming from the hidden layer by means of a subset of nodes from the same layer. This allows to considerably reduce the computational complexity compared to ELM. Experimental results based on simulation for different classification problems indicate that the proposed model achieves the same ELM performances in terms of generalization, exceeding it in speed [ABSTRACT FROM AUTHOR]

The objective of this article is to classify the 27 gestures of the Colombian sign alphabet, by means of a classifier of artificial neural networks based on electromyographic signals. The classifier was designed in four phases: Acquisition of electromyographic signals from the eight sensors of the Myo Armband handle, extraction of characteristics of the electromyographic signals using the wavelet transform of packages, training of the neural network and validation of the classification method using the cross-validation technique. For the present study, records of electromyographic signals from 13 subjects with hearing impairment were acquired. The classifier presented an average accuracy percentage of 88.4%, very similar to other classification methods presented in the literature. The classification method can be scaled to classify, in addition to the 27 gestures, the vocabulary of the Colombian sign language. [ABSTRACT FROM AUTHOR]

Artificial neural networks have been applied to climatic precipitation data, including surface sea temperatures in different areas classified as El Niño, and speed of trade winds with the purpose of modeling and predicting the climate phenomenon six months in advance to its appearance. The study was done in Piura, Peru. A preliminary analysis of the information is performed to determine the degree of correlation between variables. A model in two phases was later designed. In the first phase, neural networks using MatLab were used to model variables as time series and, in the second phase, a neural network was designed to simulate the nature of rainfall in Piura. The study shows that neural networks represents a highly reliable technique to find a pattern of precipitation and then for predicting the phenomenon with probability of 98.4% in the training step and 100% in the predicting step for the first semester of 2016. [ABSTRACT FROM AUTHOR]

A methodology based on neural networks is proposed for solving the job shop-open shop scheduling problem. The job shop scheduling problem consist of defining the best order sequence to minimize the total processing time (makespan) or other relevant variables. Several intelligent techniques have been applied to solve this kind of problem. However, when reprogramming is required these techniques present practical difficulties since they require to be restructured to find a new solution. The proposed methodology trains the network by combining the processing times at each node, through transfer functions that are multiplied by weights obtained from the algorithm of network programming. The proposed neural network obtains solutions not only for a particular problem but also for new situations without requiring a problem reconfiguration. When compared with other techniques, the obtained solution showed a superior performance ranging from 30% to 164% in terms of the makespan. [ABSTRACT FROM AUTHOR]

Gold price prediction using an artificial neural network model (ANN) is proposed in this work. The objective of the model is to predict the daily closing prices in the London market, which are taken as reference prices for the Central Bank of Colombia. Different configurations of type feed-forward ANN are considered using the dollar index DXY, the SP500 index, the daily oil price series, and the daily gold price series, as inputs to the ANN model. A set of ANN structures are trained using the historical series of data, where one portion is used for training and the other portion is used for testing (prediction). The results show good performance of the model both in the analyzed historical period and the predictions, where the best structure includes the daily price series of gold, the DXY index and the SP500 index. [ABSTRACT FROM AUTHOR]

The temperatures of atmospheric air at 850 millibars are modeled to determine their potential as an indicator of heat waves in northwest Mexico. The analysis was performed at meso-scale level but focused on the city of Mexicali, Mexico. The main variables that cause the formation of a heat wave are identified, and the modeling of air temperature at 850 milibars as an indicator of its development. This is done considering eight climate variables, using six artificial intelligence algorithms. The numerical technique of artificial neural networks showed a better performance, obtaining a regression coefficient of 0.76 with a pvalue of 0.0019. It is concluded that this non-linear model is a promising tool that could be used in a warning system of this dangerous atmospheric phenomenon. [ABSTRACT FROM AUTHOR]

A strategy has been developed for the sensorial classification of jellies made from chicken feet using neural networks based on Kohonen algorithms. These networks showed to be good tools for sensorial comparison among samples, allowing identification of the best gelatins. The sample Gelatin D, con 3.80% (w/v) of gelatin powder and 28.6 (w/v) of sugar was the best than gelatin samples from chicken feet, with its sensorial qualities varying between 6 and 8 times in hedonic scale. [ABSTRACT FROM AUTHOR]

A methodology for analyzing the roughness based on surface characteristics of the images obtained from optical and electronic microscopes, is presented. The features that describe textures and are also used to classify them derive from the Haralick descriptors, which are based on cooccurrence matrices. The primary roughness patterns are evaluated and classified according to several features which use the values of these descriptors. The values extracted from the patterns are fed to artificial neural network of the multi-layer perceptron type. It is concluded that it is possible to start implementing the control of metal parts for industrial quality control of manufactured products through this system of roughness recognition. [ABSTRACT FROM AUTHOR]

It is proposed to use the concept of ubiquitous computing for improving human-machine relations in automated homes, offices and building environments. The patterns behavior of the user is registered to use them in a context-aware, able to automatically react and allowing autoconfiguration of an automated system in an office environment. The automated mechanism which determines the state of lighting, temperature, humidity and other environment values, and how the system acts on them, are detailed. The daily habits of the user and the actual conditions of the environment are considered. Artificial Neural Networks are used to classify the state of light, heating and ventilation. The results show that the neural network technique is fully capable of recognizing more than 90% of the user patterns. [ABSTRACT FROM AUTHOR]

This paper presents a system for detecting angular position of targets, using feature extraction techniques in digital imaging and artificial neural networks. Military ships images graphically generated by three-dimensional solid modeling software are used. Several tests using artificial neural networks applied to the set of geometric features were performed. The results show the important contribution of recognition algorithms in determining the ship angular position, regardless of their distance from the observer. The results encourage future applications for tracking targets using infrared images. [ABSTRACT FROM AUTHOR]