Haruna Matsushita

Fuzzy Adaptive Resonance Theory Combining Overlapped Category in consideration of connections

Adaptive resonance theory (ART) is an unsupervised neural network. Fuzzy ART (FART) is a variation of ART, allows both binary and continuous input patterns. However, fuzzy ART has the category proliferation problem. In this study, to solve this problem, we propose a new fuzzy ART algorithm: fuzzy ART combining overlapped category in consideration of connections (C-FART). C-FART has two important features. One is to make connections between similar categories. The other is to combine overlapping categories into with connections one category. We investigate the behavior of C-FART, and compare C-FART with the conventional FART.

Network-Structured Particle Swarm Optimizer with Various Topology and Its Behaviors

This study proposes Network-Structured Particle Swarm Optimizer (NS-PSO) with various neighborhood topology. The proposed PSO has the various network topology as rectangular, hexagonal, cylinder and toroidal. We apply NS-PSO with various topology to optimization problems. We investigate their behaviors and evaluate what kind of topology would be the most appropriate for each function.

Network-Structured Particle Swarm Optimizer considering neighborhood relationships

This study proposes a new Network-Structured Particle Swarm Optimizer considering neighborhood relationships (NS-PSO). All particles of NS-PSO are connected to adjacent particles by a neighborhood relation of the 2-dimensional network. The directly connected particles share the information of their own best position. Each particle is updated depending on the neighborhood distance on the network between it and a winner, whose function value is best among all particles. Simulation results show the searching efficiency of NS-PSO.

Self-Organizing Map with False-Neighbor Degree between Neurons for Effective Self-Organization

In the real world, it is not always true that neighboring houses are physically adjacent or close to each other. in other words, “neighbors” are not always “true neighbors.” In this study, we propose a new Self-Organizing Map (SOM) algorithm, SOM with False-Neighbor degree between neurons (called FN-SOM). The behavior of FN-SOM is investigated with learning for various input data. We confirm that FN-SOM can obtain a more effective map reflecting the distribution state of input data than the conventional SOM and Growing Grid.

Batch-Learning Self-Organizing Map with Weighted Connections avoiding false-neighbor effects

This study proposes a Batch-Learning Self-Organizing Map with Weighted Connections avoiding false-neighbor effects (BL-WCSOM). We apply BL-WCSOM to several high-dimensional datasets. From results measured in terms of the quantization error, inactive neurons, the topographic error and the computation time, we confirm that BL-WCSOM obtain the effective map reflecting the distribution state of the input data using fewer neurons in less time.

Batch-Learning Self-Organizing Map with false-neighbor degree between neurons

This study proposes a batch-learning self-organizing map with false-neighbor degree between neurons (called BL-FNSOM). False-neighbor degrees are allocated between adjacent rows and adjacent columns of BL-FNSOM. The initial values of all of the false-neighbor degrees are set to zero, however, they are increased with learning, and the false-neighbor degrees act as a burden of the distance between map nodes when the weight vectors of neurons are updated. BL-FNSOM changes the neighborhood relationship more flexibly according to the situation and the shape of data although using batch learning. We apply BL-FNSOM to some input data and confirm that FN-SOM can obtain a more effective map reflecting the distribution state of input data than the conventional Batch-Learning SOM.

Competing Behavior of Two Kinds of Self-Organizing Maps and Its Application to Clustering

The Self-Organizing Map (SOM) is an unsupervised neural network introduced in the 80s by Teuvo Kohonen. In this paper, we propose a method of simultaneously using two kinds of SOM whose features are different (the nSOM method). Namely, one is distributed in the area at which input data are concentrated, and the other self-organizes the whole of the input space. The competing behavior of the two kinds of SOM for nonuniform input data is investigated. Furthermore, we show its application to clustering and confirm its efficiency by comparing with the k-means method.

Competing and accommodating behaviors of peace SOM

The self-organizing map (SOM) attracts attentions for clustering in these years. In our past study, we have proposed a method using simultaneously two kinds of SOMs whose features are different, namely, one self-organizes the area on which input data are concentrated, and the other self-organizes the whole of the input space. Further, we have applied this method to clustering of data including a lot of noises and have confirmed the efficiency. However, in order to obtain an efficient clustering performance using this method, we must determine the appropriate number of the SOMs used in the method. In this study, we propose the peace SOM (PSOM) algorithm which possesses both competing and accommodating abilities. The competing and the accommodating behaviors of PSOM are investigated with applications to clustering input data including a lot of noises. We can see that PSOM successfully extracts clusters even in the case that we do not know the number of clusters in advance

Self-Organizing Map with Weighted Connections avoiding false-neighbor effects

This study proposes the Self-Organizing Map with Weighted Connections avoiding false-neighbor effects (WC-SOM). We investigate the effectiveness of WC-SOM in comparison with the conventional SOM, Growing Grid and FN-SOM. We confirm that WC-SOM enables the most flexible self-organization among the four algorithms and can obtain the effective map reflecting the distribution state of the input data using fewer neurons.

Firefly algorithm with dynamically changing connections

This study proposes a firefly algorithm with dynamically changing connections (FA-DC). In a standard firefly algorithm (FA), a brightness of each firefly is determined by the objective function, and for any two fireflies, the less brighter one will be always attracted by the brighter one. On the other hand, the fireflies of FA-DC move depending on the connections between fireflies. Even if the brighter firefly exists, the less brighter firefly does not move toward the brighter one when there is no connection between the two fireflies. Furthermore, the connections of FA-DC changes dynamically for every iteration. This effect promotes a diversification of the solutions and avoids the solutions being trapped at local optima. We apply FA-DC to 28 optimization benchmarks from the 2013 Congress on Evolutionary computation (CEC), and we compare it with the conventional FA and the particle swarm optimization (PSO). Simulation results show that FA-DC significantly improves the optimization performance from the conventional FA although FA-DC is a simple algorithm that needs no carefully parameter tuning.