A spiking neural network that extracts frequent symbol patterns from time-series data with different symbol appearance intervals

In this research, we aim to extract frequent symbol patterns from symbol sequences. In the target time series data, the appearance timing of symbols is different. The purpose is to absorb the deviation of the appearance timing and extract frequent symbol patterns. This paper proposes a spiking neural network that extracts frequent symbol patterns. The structure of the proposed network grows automatically each time a symbol is given. By learning the network, the output unit will only fire when given a frequent symbol pattern, and the network will extract the frequent symbol patterns. As a result of a simple experiment, it was confirmed that the proposed network can extract frequent symbol patterns from the target time series data.