멀티모달 스파이킹 뉴럴 네트워크의 성능 분석

Abstract

Multimodal models have been studied as promising model that can overcome the disadvantages of the unimodal models. However, not much research has been conducted on the effectiveness of multimodality in spiking neural networks (SNNs), which have been considered a next-generation artificial neural network for their energy efficiency. Thus, in this paper, we analyzed the effectiveness through experiments on modality, model size, and noise. According to our analysis, we validated that SNNs showed greater effectiveness in multimodality than DNNs.