Claudio Moraga

Computing with Antonyms

This work tries to follow some agreements linguistic seem to have on the semantical concept of antonym, and to model by means of a membership function an antonym aP of a predicate P, whose use is known by a given μP

Multiple-Valued Logic and Artificial IntelligenceFundamentals of Fuzzy Control Revisited

This paper reviews one particular area of Artificial Intelligence, which roots may be traced back to Multiple-valued Logic: the area of fuzzy control. After an introduction based on an experimental scenario, basic cases of fuzzy control are presented and formally analyzed. Their capabilities are discussed and their constraints are explained. Finally it is shown that a parameterization of either the fuzzy sets or the connectives used to express the rules governing a fuzzy controller allows the use of new optimization methods to improve the overall performance.

Extended Kalman Filter Trained Recurrent Radial Basis Function Network in Nonlinear System Identification

We consider the recurrent radial basis function network as a model of nonlinear dynamic system. On-line parameter and structure adaptation is unified under the framework of extended Kalman filter. The ability of adaptive system to deal with high observation noise, and the generalization ability of the resulting RRBF network are demonstrated in nonlinear system identification.

Functional equivalence between S-neural networks and fuzzy models

A family of S-functions is introduced and characterized. S-functions may be used as activation functions in neural networks and allow the interpretation of the activity of the artificial neurons as fuzzy if-then rules, where the degree of satisfaction of the premises for a given input is calculated by means of the symmetric summation. These rules are appropriate to model compensating systems.

Robust Estimator for the Learning Process in Neural Networks Applied in Time Series

Artificial Neural Networks (ANN) have been used to model non-linear time series as an alternative of the ARIMA models. In this paper Feedforward Neural Networks (FANN) are used as non-linear autoregressive (NAR) models. NAR models are shown to lack robustness to innovative and additive outliers. A single outlier can ruin an entire neural network fit. Neural networks are shown to model well in regions far from outliers, this is in contrast to linear models where the entire fit is ruined. We propose a robust algorithm for NAR models that is robust to innovative and additive outliers. This algorithm is based on the generalized maximum likelihood (GM) type estimators, which shows advantages over conventional least squares methods. This sensitivity to outliers is demostrated based on a synthetic data set.

Improving the Weighted Distribution Estimation for AdaBoost Using a Novel Concurrent Approach

AdaBoost is one of the most known Ensemble approaches used in the Machine Learning literature. Several AdaBoost approaches that use Parallel processing, in order to speed up the computation in Large datasets, have been recently proposed. These approaches try to approximate the classic AdaBoost, thus sacrificing its generalization ability. In this work, we use Concurrent Computing in order to improve the Distribution Weight estimation, hence obtaining improvements in the capacity of generalization. We train in parallel in each round several weak hypotheses, and using a weighted ensemble we update the distribution weights of the following boosting rounds. Our results show that in most cases the performance of AdaBoost is improved and that the algorithm converges rapidly. We validate our proposal with 4 well-known real data sets.

Gaussian Sum Filters for Recurrent Neural Networks training

We consider the problem of recurrent neural network training as a Bayesian state estimation. The proposed algorithm uses Gaussian sum filter for nonlinear, non-Gaussian estimation of network outputs and synaptic weights. The performances of the proposed algorithm and other Bayesian filters are compared in noisy chaotic time series long-term prediction

19. Workshop "Interdisziplinäre Methoden in der Informatik"

CAMBIO system as one of the few available layout compaction systems for analog circuits has its inherent complexity in algorithm, system architecture and software structure. How to redesign this system to make it robust for the further development and maintenance will be discussed in this paper.