A general strategy for designing modular learning systems is to treat the problem as one of combining multiple models, each of which is defined over a local region of the input space. Jacobs, Jordan, Nowlan and Hinton...
详细信息
A general strategy for designing modular learning systems is to treat the problem as one of combining multiple models, each of which is defined over a local region of the input space. Jacobs, Jordan, Nowlan and Hinton introduced such a strategy with their 'mixture of experts' (ME) architecture for supervised learning. The ME architecture is closely related to the decision tree and multivariate spline algorithms. The gradient approach for training a mixture of experts architecture did not appear to take advantage of the modularity of the architecture. The Expectation-Maximization (EM) algorithm has been proposed for a mixture of experts architecture.
The paper deals with new developments on interpolating memories as the basic element of learning control and their possible applications. The paper addresses only three possible applications of learning: building a no...
详细信息
The paper deals with new developments on interpolating memories as the basic element of learning control and their possible applications. The paper addresses only three possible applications of learning: building a nonlinear process model to investigate the effects of possible control strategies;automatic formation of an inverse process model to linearize and decouple the control loop;and training of nonlinear complex controllers needing high computation effort into a memory to allow very fast control action generation.
In this paper, we prove the stability of a certain class of nonlinear discrete MIMO (Multi-Input Multi-Output) systemscontrolled by a multilayer neural net with a simple weight adaptation strategy. The proof is based...
详细信息
In this paper, we prove the stability of a certain class of nonlinear discrete MIMO (Multi-Input Multi-Output) systemscontrolled by a multilayer neural net with a simple weight adaptation strategy. The proof is based on the Lyapunov formalism. The stability statement is, however, only valid if the initial weight values are not too far from their optimal values that allow perfect model matching. We therefore propose to initialize the weights with values that solve the linear problem. This extends our previous work (Renders, 1993;Saerens, Renders & Bersini. 1993), where single-input single-output (SISO) systems were considered.
A Kohonen self organizing map was used to classify the reflectance spectra of a car body steel. The method produces more accurate results of corrosion evaluation than the spectroscopic corrosion index. Another advanta...
详细信息
A Kohonen self organizing map was used to classify the reflectance spectra of a car body steel. The method produces more accurate results of corrosion evaluation than the spectroscopic corrosion index. Another advantage of applying a Kohonen net for the classification is the more robust behavior towards spectral noise in the data.
It has long been known that the optimal equalization of communication channels is achieved using the Maximum Likelihood Sequence Estimation (MLSE). The use of nonlinear clustering has been found to extend the capabili...
详细信息
It has long been known that the optimal equalization of communication channels is achieved using the Maximum Likelihood Sequence Estimation (MLSE). The use of nonlinear clustering has been found to extend the capabilities of MLSE in the case of channel nonlinearity or time variation. However, a problem with these techniques has been either high computational complexity of uncertainty about model order. The paper provides a technique which fixes the model order at an a-priori known level and simplifies the clustering by the use of a finite polynomial interpretation which also leads to simple and stable adaptation.
The paper discusses modeling problems where there are some empirical data, and some limited system knowledge available. In such cases, first principles modeling may lead to an inaccurate model. Using black-box alterna...
详细信息
The paper discusses modeling problems where there are some empirical data, and some limited system knowledge available. In such cases, first principles modeling may lead to an inaccurate model. Using black-box alternative, not much of the system knowledge can be incorporated a priori. The authors propose an approach based on the observation that for many systems, adequate models of the behavior within small operating regimes can be found without too much difficulty, while global models, covering all possible operating conditions, tend to be very complex and difficult or expensive to build. The method constructs local models for the various operating regimes, and builds a global model by interpolating the local models.
The modelling and generalization abilities of a class of neural algorithms called neurofuzzy networks are investigated. Neurofuzzy networks try to combine the vague linguistic representation of fuzzy logic with the le...
详细信息
The modelling and generalization abilities of a class of neural algorithms called neurofuzzy networks are investigated. Neurofuzzy networks try to combine the vague linguistic representation of fuzzy logic with the learning abilities of neuralnetworks. Also discussed is the Cerebellar model articulation controller (CMAC) which is a tabular look-up table for generalizations.
The proceedings contains 7 papers on nonlinear control. Some of the topics discussed include control system synthesis and analysis, neuralnetworks, mathematical models, system stability, optimal controlsystems, unce...
详细信息
The proceedings contains 7 papers on nonlinear control. Some of the topics discussed include control system synthesis and analysis, neuralnetworks, mathematical models, system stability, optimal controlsystems, uncertainty, adaptive control, fuzzy sets, state estimation and control theory.
This paper shows the effect of introducing noise to the weight set and at the input to the neuron. The MLP investigated is tolerant to noise added at the input to the neuron and therefore could be implemented using th...
详细信息
This paper shows the effect of introducing noise to the weight set and at the input to the neuron. The MLP investigated is tolerant to noise added at the input to the neuron and therefore could be implemented using the PWM neural network with the RC time constant set close to the PWM period.
暂无评论