Catálogo de publicaciones - libros

The problem of robust stability for a class of uncertain bidirectional associative memory neural networks with time delays is investigated in this paper. A more general Lyapunov-Krasovskii functional is proposed to derive a less conservative robust stability condition within the framework of linear matrix inequalities. A numerical example is given to illustrate the effectiveness of the proposed method.

The map-coloring problem is a well known combinatorial optimization problem which frequently appears in mathematics, graph theory and artificial intelligence. This paper presents a study into the performance of some binary Hopfield networks with discrete dynamics for this classic problem. A number of instances have been simulated to demonstrate that only the proposed binary model provides optimal solutions. In addition, for large-scale maps an algorithm is presented to improve the local minima of the network by solving gradually growing submaps of the considered map. Simulation results for several n-region 4-color maps showed that the proposed neural algorithm converged to a correct colouring from at least 90% of initial states without the fine-tuning of parameters required in another Hopfield models.

Currently foot pathologies, like cave and flat foot, are detected by an human expert who interprets a footprint image. The lack of trained personal to carry out massive first screening detection campaigns precludes the routinary diagnostic of these pathologies. This work presents a novel automatic system, based on Neural Networks (NN), for foot pathologies detection. In order to improve the efficiency of the neural network training algorithm, we propose the use of principal components analysis to reduce the number of inputs to the NN. The results obtained with this system demonstrate the feasibility of building automatic diagnosis systems based on the foot image. These systems are very valuable in remote areas and can be also used for massive first screening health campaigns.

Many metallurgical processes are complex and due to hostile environment it is difficult to carry out reliable measurement of their internal state, but the demands on high productivity and consideration of environmental issues require that the processes still be strictly controlled. Due to the complexity and non-ideality of the processes, it is often not feasible to develop mechanistic models. An alternative is to use neural networks as black-box models, built on historical process data. The selection of relevant inputs and appropriate network structure are still problematic issues. The present work addresses these two problems in the modeling of the hot metal silicon content in the blast furnace. An algorithm is applied to find relevant inputs and their time lags, as well as a proper network size, by pruning a large network. The resulting models exhibit good prediction capabilities and the inputs and time lags detected are in good agreement with practical metallurgical knowledge.

This paper proposed a novel dynamic system which utilizes Ridge Polynomial Neural Networks for the prediction of the exchange rate time series. We performed a set of simulations covering three uni-variate exchange rate signals which are; the JP/EU, JP/UK, and JP/US time series. The forecasting performance of the novel Dynamic Ridge Polynomial Neural Network is compared with the performance of the Multilayer Perceptron and the feedforward Ridge Polynomial Neural Network. The simulation results indicated that the proposed network demonstrated advantages in capturing noisy movement in the exchange rate signals with a higher profit return.

In this paper a neural system for daily forecasting of electric power load in Poland is presented. Basing on the simplest neural architecture - a multi-layer perceptron - more and more complex system is built step by step. A committee rule-aided hierarchical system consisting of modular ANNs is obtained as a result. The forecasting mean absolute percentage error (MAPE) of the most effective system is about 1.1%.

This paper is concerned with the approximation problem of the SO-3 jet engine turbine and compressor characteristics. Topology selection of multilayer feedforward artificial neural networks is investigated. Neural models are compared with Takagi-Sugeno fuzzy models in terms of approximation accuracy and complexity.

This paper proposes a speech enhancement system based on an auditory system for noise reduction in speech that is degraded by background noises. Accordingly, the proposed system adjusts frame by frame the coefficients for both lateral inhibition and amplitude component according to the detected sections for each input frame, then reduces the noise signal using a time-delay neural network. Based on measuring signal-to-noise ratios, experiments confirm that the proposed system is effective for speech that is degraded by various noises.

Automatic pattern recognition is a very important task in many applications such as image segmentation, object detection, etc. This work aims to find a new approach to automatically recognize patterns such as 3D objects and handwritten digits based on a database using General Regression Neural Networks (GRNN). The designed system can be used for both 3D object recognition from 2D poses of the object and handwritten digit recognition applications. The system does not require any preprocessing and feature extraction stage before the recognition. Simulation results show that pattern recognition by GRNN improves the recognition rate considerably in comparison to other neural network structures and has shown better recognition rates and much faster training times than that of Radial Basis Function and Multilayer Perceptron networks for the same applications.

A novel approach to real-time string filtering of large databases is presented. The proposed approach is based on a combination of artificial neural networks and operates in two stages. The first stage employs a self-organizing map for performing approximate string matching and retrieving those strings of the database which are similar to (i.e. assigned to the same SOM node as) the query string. The second stage employs a harmony theory network for comparing the previously retrieved strings in parallel with the query string and determining whether an exact match exists. The experimental results demonstrate accurate, fast and database-size independent string filtering which is robust to database modifications. The proposed approach is put forward for general-purpose (directory, catalogue and glossary search) and Internet (e-mail blocking, intrusion detection systems, URL and username classification) applications.