Hiroaki Hara
Department of Science Engineering, Faculty of Engineering, Tohoku University
Keywords: Kanji characters, memory, random walks.
Abstract. A mathematical model of memory for pattern recognition in a neural network system is proposed. Categories in an information source, the Kanji (Chinese) characters, are transmitted to a receptor and they are encoded by a set of pulses in the neural network. Storages of memory are assumed to be expressed by a distribution of membrane potentials of neurons. The model describes the storages of memory for the pattern recognition by multi-dimensional stochastic processes of neurons. Specifically, when the Kanji characters are recognized in the order of making the strokes, it is found that the recall procedure shows a profile similar to the serial position curve observed in the serial recall procedures for the items.