Abstract
Recently, artificial neural networks which mimic more precise the structure of the brain compared with conventional artificial neural networks have been studied actively, e.g. Deep Neural Networks. Self-Organizing Map(SOM) is a well-known model of visual cortex of a brain. Neurons in real visual cortex of a brain has horizontal connections each other. On the other hand, the basic SOM has no connections among neurons in competitive layer. In this research, we propose to introduce horizontal connections of neurons in competitive layer of the basic SOM to represent exactly the real visual cortex. Our algorithm define the representative neuron and connections among that neurons, and it detects similarity among classes of input data.
