Eduardo Gómez Sánchez

MicroARTMAP: use of mutual information for category reduction in fuzzy ARTMAP

A new architecture, called MicroARTMAP, is proposed to impact the category proliferation problem present in Fuzzy ARTMAP. It handles probabilistic information through the optimization of the mutual information between the input and output spaces, but allowing a small training error, thus avoiding overfitting. While reducing the number of categories used by Fuzzy ARTMAP, it holds several desirable properties, such as a correct treatment of exceptions and a fast algorithm, as opposed to other approaches like BARTMAP. In addition, it is shown that MicroARTMAP is less sensitive than Fuzzy ARTMAP with respect to the the pattern presentation order, and that it degrades less if the training set is noisy.

Learning from noisy information in FasArt and FasBack neuro-fuzzy systems

Neuro-fuzzy systems have been in the focus of recent research as a solution to jointly exploit the main features of fuzzy logic systems and neural networks. Within the application literature, neuro-fuzzy systems can be found as methods for function identification. This approach is supported by theorems that guarantee the possibility of representing arbitrary functions by fuzzy systems. However, due to the fact that real data are often noisy, generation of accurate identifiers is presented as an important problem. Within the Adaptive Resonance Theory (ART), PROBART architecture has been proposed as a solution to this problem. After a detailed comparison of these architectures based on their design principles, the FasArt and FasBack models are proposed. They are neuro-fuzzy identifiers that offer a dual interpretation, as fuzzy logic systems or neural networks. FasArt and FasBack can be trained on noisy data without need of change in their structure or data preprocessing. In the simulation work, a comparative study is carried out on the performances of Fuzzy ARTMAP, PROBART, FasArt and FasBack, focusing on prediction error and network complexity. Results show that FasArt and FasBack clearly enhance the performance of other models in this important problem.

A neuro-fuzzy system that uses distributed learning for compact rule set generation

ARTMAP based architectures have several desirable properties that make them very suitable for pattern classification problems. However, they suffer from category proliferation. Distributed coding has been proposed as a solution for memory compression and the dARTMAP neural network has been introduced as a modification of the fuzzy ARTMAP that, due to distributed learning, achieves code compression while fast stable learning is retained. A critical analysis of dARTMAP architecture and performance in pattern recognition problems is presented, concluding that distributed learning excels the original fuzzy ARTMAP only under certain geometrical configurations of the output classes, or in the presence of noise in the training set. A new architecture called dFasArt is presented, introducing distributed learning into the FasArt neuro-fuzzy system, which is more suitable for identification tasks, showing that the advantages of distributed code can be extended to other neural architectures. Experimental results show that dFasArt performs similarly to dARTMAP in classification tasks, while being less sensitive to pattern presentation order.

FLAS: a fuzzy linear adaptive system for identification of non-linear noisy functions

FLAS (fuzzy linear adaptive system) is a self-organizing fuzzy system for non-linear function identification, that uses a learning method based on clustering to generate fuzzy rules and tune their parameters. This method reduces the influence of pattern presentation order, permits building prototypes with physical meaning, allows measuring the importance of each variable, and therefore reduces the influence of noise. FLAS fuzzy membership functions are defined as barycentric coordinates in a simplex, yielding equivalence between Mandami and Takagi-Sugeno defuzzification methods. This allows FLAS to make piecewise linear interpolation and thus facilitates a rule fusion procedure. In simulations done for noisy non-linear function identification tasks, FLAS showed better results than other comparative systems yielding smaller identification error and number of rules. In the difficult task of bioprocesses variable identification FLAS also outperforms other systems. FLAS theoretical features and good identification performance provide good expectations for its implementation within different model based controllers.

Entorno de simulación distribuida de redes basado en la nube computacional

Las simulaciones de barridos de parametros tienen un gran potencial en el estudio de redes telematicas, especialmente en contextos docentes. Sin embargo, el elevado tiempo necesario habitualmente para completar este tipo de simulaciones es una limitacion importante para su uso. En este articulo se propone DNSE3, un entorno que permite la ejecucion distribuida de tareas de simulacion en el simulador ns-3 dentro de un entorno de nube computacional, a traves de una arquitectura de servicios RESTful. El sistema se ha disenado para ser autoescalable, aprovisionando y liberando dinamicamente recursos de la nube computacional en funcion de la carga de simulaciones demandada, y garantizando un reparto equitativo de los recursos entre los distintos usuarios. Ademas, DNSE3 se ha implementado reutilizando servicios presentes en \emph{middlewares} de nube populares, y ha sido evaluado mediante pruebas sinteticas. La implementacion de DNSE3 ha demostrando un correcto comportamiento funcional y un rendimiento considerablemente superior a otras alternativas cuando el numero de simulaciones es muy elevado.