Catálogo de publicaciones - libros

Selective attention to visual-spatial stimuli causes decrements of power in alpha band and increments in beta. For steady-state visual evoked potentials (SSVEP) selective attention affects electroencephalogram (EEG) recordings, modulating the power in the range 8-27 Hz. The same behaviour can be seen for auditory stimuli as well, although for auditory steady-state response (ASSR), it is not fully confirmed yet. The design of selective attention based braincomputer interfaces (BCIs) has two major advantages: First, no much training is needed. Second, if properly designed, a steady-state response corresponding to spectral peaks can be elicited, easy to filter and classify. In this paper we study the behaviour of Kohonen Maps as feature selector for a selective attention to auditory stimuli based BCI system.

Nature has been a continuous source of inspiration for many successful techniques, algorithms and computational metaphors. We outline such a inspiration here, in the context of bio-inspired congestion control (BICC) algorithms. In this paper a realistic predator-prey model is mapped to the Internet congestion control mechanism. This mapping leads to a bio-inspired congestion control scheme. Dynamic and equilibrium properties of developed algorithm are good enough according to the simulation results.

In this work we present a new proposal in order to model the probability distribution in the estimation of distribution algorithms. This approach is based on using dependency networks [1] instead of Bayesian networks or simpler models in which structure is limited. Dependency networks are probabilistic graphical models similar to Bayesian networks, but with a significant difference: they allow directed cycles in the graph. This difference can be an important advantage because of two main reasons. First, in some real problems cyclic relationships appear between variables an this fact cannot be represented in a Bayesian network. Secondly, dependency networks can be built easily due to the fact that there is no need to check the existence of cycles as in a Bayesian network.

In this paper we propose to use a general (multivariate) model in order to deal with a richer representation, however, in this initial approach to the problem we also propose to constraint the construction phase in order to use only bivariate statistics. The algorithm is compared with classical approaches with the same complexity order, i.e. bivariate models as chains and trees.

This article proposes a context-free grammar to be used in grammar-guided genetic programming systems to automatically design feed-forward neural architectures. This grammar has three important features. The sentences that belong to the grammar are binary strings that directly encode all the valid neural architectures only. This rules out the appearance of illegal points in the search space. Second, the grammar has the property of being ambiguous and semantically redundant. Therefore, there are alternative ways of reaching the optimum. Third, the grammar starts by generating small networks. This way it can efficiently adapt to the complexity of the problem to be solved. From the results, it is clear that these three properties are beneficial to the convergence process of the grammar-guided genetic programming system.

The paper describes a methodology for constructing a possible combination of different basis functions (sigmoidal and product) for the hidden layer of a feed forward neural network, where the architecture, weights and node typology are learned based on evolutionary programming. This methodology is tested using simulated Gaussian data set classification problems with different linear correlations between input variables and different variances. It was found that combined basis functions are the more accurate for classification than pure sigmoidal or product-unit models. Combined basis functions present competitive results which are obtained using linear discriminant analysis, the best classification methodology for Gaussian data sets.

In this article, a methodology to obtain the Feasible Parameter Set () and a nominal model in a non-linear robust identification problem is presented. Several norms are taken into account simultaneously to define the which improves the model quality but, as counterpart, it increases the optimization problem complexity. To determine the a multimodal optimization problem with an infinite number of minima, which constitute the , is presented and a special evolutionary algorithm (−GA) is used to characterize it. Finally, an application to a thermal process identification, where ||·|| and ||·|| norms have been considered simultaneously, is presented to illustrate the technique.

We consider the numerical optimization of dynamic molecular alignment by shaped femtosecond laser pulses. We study a simplified model of this problem, which allows the full physical investigation of the optimal solutions. By using specific variants of Derandomized Evolution Strategies, subject to parameterizations which are known to be superior for this problem, the numerical results reveal different for the different optimization procedures. These results are strong both from the algorithmic as well as from the physics perspectives. This shows that techniques can be used to derive new insights into .

In this work we use simulated evolution to corroborate the adaptability of the natural genetic code. An adapted genetic algorithm searches for optimal hypothetical codes. The adaptability is measured as the average variation of the hydrophobicity that experiment the encoded amino acids when errors or mutations are presented in the codons of the hypothetical codes. Different types of mutations and base position mutation probabilities are considered in this study.

The new learning classifier system is described to classify real-valued data. The approach applies the continous-valued context-free grammar based system GCS. In order to handle effectively, the terminal rules have been replaced by the so-called environment probing rules. The rGCS model has been tested on the checkerboard problem.

This work presents the results on applying a genetic approach for solving the Dial-A-Ride Problem (DARP). The problem consists of assigning and scheduling a set of user transport requests to a fleet of available vehicles in the most efficient way according to a given objective function. The literature offers different heuristics for solving DARP, a well known NP-hard problem, which range from traditional insertion and clustering algorithms to soft computing techniques. On the other hand, the approach through Genetic Algorithms (GA) has been experienced in problems of combinatorial optimization. We present our experience and results of a study to develop and test different GAs in the aim of finding an appropriate encoding and configuration, specifically for the DARP problem with time windows.