Extraction of frequent sub-sequences from sequential input using spiking neuron

Abstract

We propose a method to extract frequent sub-sequences from sequential input. Especially, we aim to extract three or more symbols from sequential input. In our former research, we could extract only two symbols. To extract three or more symbols, we propose a learning method and structure of self-organizing spiking neural network. The learning method is based on STDP rule. An output-layer of neural network has many unit groups. We confirmed the proposed method could extract frequent sub-sequence.