Showing papers by "Ignacio Rojas published in 1999"

An Accurate Measure for Multilayer Perceptron Tolerance to Weight Deviations

Abstract: The inherent fault tolerance of artificial neural networks (ANNs) is usually assumed, but several authors have claimed that ANNs are not always fault tolerant and have demonstrated the need to evaluate their robustness by quantitative measures. For this purpose, various alternatives have been proposed. In this paper we show the direct relation between the mean square error (MSE) and the statistical sensitivity to weight deviations, defining a measure of tolerance based on statistical sentitivity that we have called Mean Square Sensitivity (MSS); this allows us to predict accurately the degradation of the MSE when the weight values change and so constitutes a useful parameter for choosing between different configurations of MLPs. The experimental results obtained for different MLPs are shown and demonstrate the validity of our model.

A new methodology to obtain fuzzy systems autonomously from training data

Abstract: This paper presents an approach to obtain a fuzzy system automatically from numerical data. The identification of the fuzzy system structure (number of rules and membership functions in each input variable) and the optimization of the parameters defining it are performed jointly. Starting from an initially simple fuzzy system, the numbers of membership functions in the input domain and of rules are adapted in order to reduce the approximation error. This method has the advantage that it does not require the human expert's assistance since the input-output characteristics of the fuzzy system and its structure are obtained from the training examples.

A new approach for the design of fuzzy controllers in real time

Abstract: This paper presents a new methodology to achieve real time self tuning and self-learning in fuzzy controllers. The advantage of this approach is that it only requires qualitative information about the plant to be controlled, in terms of the monotony presented by the output with respect to the control signal and delays of the plant. Thus, it is capable of controlling highly nonlinear systems, in a pseudo-optimum way, even when these are time variable. Control is achieved by means of two auxiliary systems: the first one is responsible for adapting the consequences of the main controller to minimize the error arising at the plant output, while the second auxiliary system compiles real input/output data obtained from the plant. The system then learns from these data, adapting both the consequences of the rules and the parameters that define the membership functions, taking into account, not the current state of the plant but rather the global identification performed.

Parallel combinatorial optimization with evolutionary cooperation between processors

Abstract: An evolutionary computation approach is used to learn online the rules that allow the processors in a parallel platform to cooperate by interchanging the local optima that they find while they concurrently explore different zones of the solution space. The cooperation of processors can greatly benefit the resolution of combinatorial optimization problems by decreasing their runtimes, by increasing the quality of the solutions obtained, or both. Moreover, as parallel computers are more and more accessible, the application of parallel processing to solve these problems becomes a practical and interesting alternative. As an example, a parallel optimization algorithm based on Boltzmann Machine has been used for a detailed description and evaluation of the proposed cooperation approach.

A new measurement of noise immunity and generalization ability for MLPs.

Abstract: This paper shows a quantitative relation between the regularization techniques, the generalization ability, and the sensitivity of the Multilayer Perceptron (MLP) to input noise. Although many studies about these topics have been presented, in most cases only one of the problems is addressed, and only experimentally obtained evidence is provided to illustrate some kind of correlation between generalization, noise immunity and the use of regularization techniques to obtain a set of weights after training that provides the corresponding MLP with generalization ability and noise immunity. Here, a new measurement of noise immunity for a MLP is presented. This measurement, which is termed Mean Squared Sensitivity (MSS), explicitly evaluates the Mean Squared Error (MSE) degradation of a MLP when it is perturbed by input noise, and can be computed from the statistical sensitivities (previously proposed) of the output neurons. The MSS provides an accurate evaluation of the MLP performance loss when its inputs are perturbed by noise and can also be considered a measurement of the smoothness of the error surface with respect to the inputs. Thus, as the MSS can be used to evaluate the noise immunity or the generalization ability, it gives a criterion to select among different weight configurations that present a similar MSE after training.

Stimulus correlation and adaptive motion detection using spiking neurons

Abstract: Stimulus correlation and adaptive movement detection, among other tasks can be performed with VLSI general-purpose neurons that have controllable steady and transient responses. This paper presents experimental results of simple neural primitives based on the CMOS neuron approach described in [11]. Stimulus correlation experiments illustrate the well defined behavior of the CMOS approach. This basic primitive is used to implement motion detectors with adaptive capabilities that enable it to work efficiently in a wide velocity range.

An accurate measure for multilayer perceptron tolerance to additive weight deviations

Abstract: The inherent fault tolerance of artificial neural networks (ANNs) is usually assumed, but several authors have claimed that ANNs are not always fault tolerant and have demonstrated the need to evaluate their robustness by quantitative measures. For this purpose, various alternatives have been proposed. In this paper we show the direct relation between the mean square error (MSE) and the statistical sensitivity to weight deviations, defining a measure of tolerance based on statistical sentitivity that we have called Mean Square Sensitivity (MSS); this allows us to predict accurately the degradation of the MSE when the weight values change and so constitutes a useful parameter for choosing between different configurations of MLPs. The experimental results obtained for different MLPs are shown and demonstrate the validity of our model.

Implementation of adaptable and hierarchical fuzzy T-norm

Abstract: A new configurable multiple-input fuzzy T-norm circuit is proposed, based on a monotonically decreasing function for the construction of the T-norm. The operator can dynamically adapt its behaviour (i.e. its behaviour can lie between the minimum, the product, or other parametric operators), and allows the use of hierarchical variables and rules.

New design for an online adaptive fuzzy controller

Abstract: Substantial developments in optimizing control methods for different purposes have been made in the field of fuzzy control in recent years. However, most of them are based on a known system model, whereas in practice such models are not usually available due to the complexity of the plant to be controlled. In this paper, an adaptive fuzzy controller optimizes the altitude control of a helicopter, in a way that does not depend on a process model. The algorithm does not need a mathematical model of the plant or its approximation in the form of a Jacobian matrix. Neither is it necessary to know the desired response at each instant of time, nor is there a need for sample data of the plant qualitative knowledge, and auxiliary fuzzy controllers enable the main controller to accomplish its task in real time.

Sequential Learning Algorithm for PG-RBF Network Using Regression Weights for Time Series Prediction

Abstract: We propose a modified radial basis function network (RBF) in which the main characteristics are that: a) the gaussian function is modified using pseudo-gaussian (PG) in which two scaling parameters σ are introduced; b) the activation of the hidden neurons is normalized c) instead of using a single parameter for the output weights, these are functions of the input variables; d) a sequential learning algorithm is presented to adapt the structure of the network, in which it is possible to create a new hidden unit and also to detect and remove inactive units. It is shown that the modified PG-RBF can reduce the number of didden units significantly compared with the classical RBF network. The feasibility of the resulting algorithm for the neural network to evolve and learn is demonstrated by predicting time series.

A VLSI approach for spike timing coding

Abstract: The paper describes a VLSI viable integrate-and-fire neuron model with an easily controllable firing threshold that can be used to induce synchronization processes. The circuits are intended to exploit both rate and spike time coding schemes, taking advantage of these synchronization processes to accelerate processing tasks. In this way the temporal domain can be exploited in neural computation architectures. A simple neural structure is also discussed, providing simulation results to illustrate how these time coded signals can be combined to perform a simple processing task such as coherent input detection.