An Adaptive Multiobjective Approach to Evolving ART Architectures

Abstract

Abstract-In this paper, we present the evolution of adaptive resonance theory (ART) neural network architectures (classifiers) using a multiobjective optimization approach. In particular, we propose the use of a multiobjective evolutionary approach to simultaneously evolve the weights and the topology of three well-known ART architectures; fuzzy ARTMAP (FAM), ellipsoidal ARTMAP (EAM), and Gaussian ARTMAP (GAM). We refer to the resulting architectures as MO-GFAM, MO-GEAM, and MO-GGAM, and collectively as MO-GART. The major advantage of MO-GART is that it produces a number of solutions for the classification problem at hand that have different levels of merit [accuracy on unseen data (generalization) and size (number of categories created)]. MO-GART is shown to be more elegant (does not require user intervention to define the network parameters), more effective (of better accuracy and smaller size), and more efficient (faster to produce the solution networks) than other ART neural network architectures that have appeared in the literature. Furthermore, MO-GART is shown to be competitive with other popular classifiers, such as classification and regression tree (CART) and support vector machines (SVMs).

Effective connectivity at synaptic level in humans: a review and future prospects

“…There are also adaptive neural network models (Kaylani et al 2010;Vigdor and Lerner 2007) based on fuzzy ARTMAPs (Carpenter et al 1992). ARTMAP (Carpenter et al 1991) is a supervised self-organizing neural network model based on adaptive resonance theory (ART).…”

Multi-objective memetic evolution of ART-based classifiers

“…Of them, Iris, Glass, and Pima were also used in [7], while Abalone was used in [8]. None of them have missing attributes, and, together, they present a typical variety of classification problems.…”

“…A recently-published evolutionary algorithm that trains ART-based classifiers is Multi-Objective Genetic ART (MO-GART) ( [6], [7], [8]). Research on Adaptive Resonance Theory (ART) neural networks has been active over the last two decades and several architectures have been proposed over the years.…”

“…Having access to such a set of models is an advantage for the designer, as it facilitates the identification of models that demonstrate the best generalization performance for minimum structural complexity. In [8] it is shown that MO-GART rapidly converges to a set of high-quality models, whose champion networks' generalization performance rivals, if not exceeds, the performance of Support Vector Machine or CART decision tree classifiers.…”

Machine learning in complex networks: modeling, analysis, and applications

“…Early developments include the famous Self-Organizing Map (SOM -Self-organizing Map) (Kohonen, 1990), Differential Competitive Learning (Kosko, 1991), and Adaptive Resonance Theory (ART -Adaptive Resonance Theory) (Carpenter and Grossberg, 1987;. From then on, many competitive learning neural networks have been proposed (Allinson et al, 2001;Amorim et al, 2007;Athinarayanan et al, 2002;Jain et al, 2010;Kaylani et al, 2010;López-Rubio et al, 2009;Lu and Ip, 2009;Meyer-Bäse and Thümm-ler, 2008;Príncipe and Miikkulainen, 2009;Tan et al, 2008) and a wide range of applications has been considered. Some of these include data clustering, data visualization, pattern recognition, and image processing (Bacciu and Starita, 2008;Chen et al, 2005;Deboeck and Kohonen, 2010;do Rêgo et al, 2010;Liu et al, 2008;Wang et al, 2009;Xu and II, 2005).…”