Catálogo de publicaciones - libros

In this paper we present application of various versions of the particle swarm optimization method (PSO) in the process of generation of multiple-classifier systems (MCS). While some of the investigated optimisation problems naturally lend themselves to the type of optimisation for which PSO is most suitable we present some other applications requiring non-standard representation of the particles as well as handling of constraints in the optimisation process. In the most typical optimisation case the continuous version of PSO has been successfully applied for the optimization of a soft-linear combiner. On the other hand, one of the adapted binary versions of PSO has been shown to work well in the case of multi-stage organization of majority voting (MOMV), where the search dimension is high and the local search techniques can often get stuck in local optima. All three presented PSO based methods have been tested and compared to each other and to forward search and stochastic hillclimber for five real-world non-trivial datasets.

The design of Radial Basis Function Neural Networks (RBFNNs) still remains as a difficult task when they are applied to classification or to regression problems. The difficulty arises when the parameters that define an RBFNN have to be set, these are: the number of RBFs, the position of their centers and the length of their radii. Another issue that has to be faced when applying these models to real world applications is to select the variables that the RBFNN will use as inputs. The literature presents several methodologies to perform these two tasks separately, however, due to the intrinsic parallelism of the genetic algorithms, a parallel implementation will allow the algorithm proposed in this paper to evolve solutions for both problems at the same time. The parallelization of the algorithm not only consists in the evolution of the two problems but in the specialization of the crossover and mutation operators in order to evolve the different elements to be optimized when designing RBFNNs. The subjacent Genetic Algorithm is the Non-Sorting Dominated Genetic Algorithm II (NSGA-II) that helps to keep a balance between the size of the network and its approximation accuracy in order to avoid overtraining networks. Another of the novelties of the proposed algorithm is the incorporation of local search algorithms in three stages of the algorithm: initialization of the population, evolution of the individuals, and final optimization of the Pareto front. The initialization of the individuals is performed hybridizing clustering techniques with the Mutual Information theory (MI) to select the input variables. As the experiment will show, the synergy of the different paradigms and techniques combined by the presented algorithm allow to obtain very accurate models using the most significant input variables.

In this paper we propose a classification method based on a special class of feed-forward neural network, namely product-unit neural networks, and on a dynamic version of a hybrid evolutionary neural network algorithm. The method combines an evolutionary algorithm, a clustering process, and a local search procedure, where the clustering process and the local search are only applied at specific stages of the evolutionary process. Our results with the product-unit models and the evolutionary approach show a very interesting performance in terms of classification accuracy, yielding a state-of-the-art performance.

The simultaneous topology optimization and training of neu- ral networks is a problem widely studied in the last years, specially for feedforward models. In the case of recurrent neural networks, the existing proposals attempt to only optimize the number of hidden units, since the problem of topology optimization is more difficult due to the feedback connections in the network structure. In this work, we make a study of the effects and difficulties for the optimization of network connections, hidden neurons and network training for dynamical recurrent models. In the experimental section , the proposal is tested in time series prediction problems.

One of the most important problems to be solved in Direct Sequence Spread Spectrum systems is to achieve a robust acquisition of the pseudonoise sequence. In time-varying environments this fact becomes even more important because acquisition and tracking performance can heavily degrade communication reliability. In this work a new multiresolutive acquisition system with a fuzzy logic estimator is proposed. The fuzzy logic estimation improves the accuracy of the acquisition stage compared to the results for the stability controller, through the estimation of the probability of being acquired, and the signal to noise ratio in the channel.

In this contribution we carry out an analysis of the Fuzzy Reasoning Methods for Fuzzy Rule Based Classification Systems in the framework of balanced and imbalanced data-sets with different degrees of imbalance. We analyze the behaviour of the Fuzzy Rule Based Classification Systems searching for the best type of Fuzzy Reasoning Method in each case, also studying the cooperation of some pre-processing methods of instances for imbalanced data-sets. To do so we use a fuzzy rule learning method that extends the well-known Wang and Mendel algorithm to classification problems.

The results obtained show the necessity to apply an instance pre-processing step and the differences for the most appropriate Fuzzy Reasoning Method in balanced and imbalanced data-sets, concluding that the choice of the Fuzzy Reasoning Method depends on the degree of imbalance, being the most adequate the use of the Winning Rule for high imbalanced data-sets and the Additive Combination method for the remaining data-sets.

In this paper a fuzzy system for automatically assessing the students’ teamwork performance is presented. The main goal of this work is to guarantee an equitable assessment of students’ teamwork throughout the course and across the lecturers of the same subject when subjective criteria are considered. The proposed fuzzy system (i) is designed by using a methodology based on a trade-off between accuracy and intelligibility, and (ii) uses as input linguistic variables a set of four statistical-based parameters, computed from real individual and group marks, which have been subjectively and objectively validated. Finally, the fuzzy system is described and validated experimentally.

Network-based applications are essential to providing intelligence to complex electromechanical processes through networked control systems (NCS). The focus of this paper is the design and application of fuzzy logic control for a type of NCSs. In order to assess its feasibility, a networked control system for high-performance milling process, a type of a complex electromechanical process, is implemented on a multi-point interface (MPI) bus, a proprietary programming interface port for peer-to-peer communications that resembles the PROFIBUS protocol. The manipulated input variable the feed rate as well as the control output variable, spindle torque, are transmitted through this network. A simple computational procedure can run remotely as an optimization function without requiring additional hardware. The results demonstrate that the Fuzzy Logic-based strategy provides accuracy, and adequate machining production time thus increasing the metal removal rate.

This paper presents a new methodology for the adjustment of fuzzy inference systems, which uses technique based on error back-propagation method. The free parameters of the fuzzy inference system, such as its intrinsic parameters of the membership function and the weights of the inference rules, are automatically adjusted. This methodology is interesting, not only for the results presented and obtained through computer simulations, but also for its generality concerning to the kind of fuzzy inference system used. Therefore, this methodology is expandable either to the Mandani architecture or also to that suggested by Takagi-Sugeno. The validation of the presented methodology is accomplished through estimation of time series and by a mathematical modeling problem. More specifically, the Mackey-Glass chaotic time series is used for the validation of the proposed methodology.

In this paper we present a method for the automatic selection of input variables and some previous parameters, such as number and type of membership functions, in an Adaptive Neuro Fuzzy Inference System (ANFIS) using a Genetic Algorithm with a new fitness function. Both of them constitute a design scheme that we will use for modeling the perception of textures in Digital I-mages. Some examples are presented, training ANFIS with this scheme for mo-deling the following visual textures: coarseness, directionality and regularity.