-Function approximation has been used extensively with rein forcement learning, even though theoretical support was based mainly on tabular representations. This paper proposes an actor-critic structure following the existing convergence proofs as much as possible. The actor and critic modules employ an adaptive neuro-fuzzy architecture based on fuzzy ARTMAP concepts and gradient descent. Resul...

A new supervised neural network architecture has been introduced, called Supervised ART-I. It has the accuracy of Fuzzy ARTMAP in classifying of both binary and analog arbitrary multi-valued input patterns. However, it is quicker in learning and classifying, has fewer parameters and requires less memory, due to its simple architecture. The Supervised ART-I has been built from a single Fuzzy ART...

The patterns which are presented to a Fuzzy ARTmap network should be preprocessed in such a way that the data are of appropriate clearance. In order to decrease the degree of similarity between the normalized binary patterns, a new method is proposed in which the binary patterns are converted to fuzzy (analogue) patterns. This is done by weighting and normalizing the values, determining the bin...

A measure of s k c e s s for any learning algorithm is how u s e ful it is in a variety of learning situations. Those learning algorithms that support universal function approximation can theoretically h e applied to a very large and interesting class of learning problems. Many kinds of neural network architectures have already been shown to support universal approximation. In this paper, we wi...

Distributed coding at the hidden layer of a multi–layer perceptron (MLP) endows the network with memory compression and noise tolerance capabilities. However, an MLP typically requires slow off–line learning to avoid catastrophic forgetting in an open input environment. An adaptive resonance theory (ART) model is designed to guarantee stable memories even with fast on–line learning. However, AR...

This paper focuses on the evolution of Fuzzy ARTMAP neural network classifiers, using genetic algorithms, with the objective of improving generalization performance (classification accuracy of the ART network on unseen test data) and alleviating the ART category proliferation problem (the problem of creating more than necessary ART network categories to solve a classification problem). We refer...

In order to reduce the effect of the category proliferation phenomenon in Fuzzy ARTMAP (FAM) and in ellipsoidal ARTMAP (EAM) architectures, The genetic algorithms were used to evolve networks of both architectures called GFAM and GEAM [3][4]. The results were very promising and the category proliferation (CP) phenomenon was minimized in most of the experiments, however, the author noticed that ...

This paper examines the generalization characteristic of Gaussian ARTMAP (GAM) neural network in classification tasks. GAM performance for classification during training and testing is evaluated using the k-folds cross validation technique. A comparison is also done between GAM and Fuzzy ARTMAP (FAM) neural network. It is found that GAM performs better (98-99%) when compared to FAM (79-82%) usi...

Remote sensing images contain a lot of mixed image pixels, but it is difficult to classify these pixels. If the number of pixel’s end-member is regarded as unchangeable, the traditional pixel unmixing algorithm cannot get a good result. In this paper we develop a new method of selective end-members for pixel unmixing based on the fuzzy ARTMAP neural network, which firstly compares the pixel’s s...

This paper shows how knowledge, in the form of fuzzy rules, can be derived from a. self-organizing supervised learning neural network called fuzzy ARTMAP. Rule extraction proceeds in two stages: pruning removes those recognition nodes whose confidence index falls below a selected threshold; and quantization of continuous learned weights allows the final system state to be translated into a usab...