Creating smaller neural networks by applying seismic waveform processing

Abstract

The purpose of this paper is to confirm the applicability of a new method for creating smaller neural networks. First, the outline and detail of the new method is explained. This method is an application of seismic waveform processing to eliminate noise that does not contribute to the output. Then, the proposed method is applied to a trained neural network as an example. Finally, the structure of the resulting neural network is compared with that of the original neural network, and the applicability of the proposed method is examined.