ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper A Structure and Generation Method of Lookup Table for Lightweight Convolutional Neural Network Design
Cited - time in scopus Share share facebook twitter linkedin kakaostory
Authors
Suwoong Lee, Seungjae Lee, Jong Gook Ko, Keundong Lee, Weon-Geun OH
Issue Date
2019-02
Citation
International Workshop on Frontiers of Computer Vision (IW-FCV) 2019, pp.1-3
Language
English
Type
Conference Paper
Abstract
To design a convolutional neural network with constrained computational resources, estimating latency of intermediate or final neural network is crucial part. Lots of previous approaches estimated the latency by indirect metric like MACs, but recent researches like NetAdapt [6] stated that direct metric should be used because indirect metric tends to have high error especially in Mobile environment. In this paper, we propose the structure and generation method of a lookup table for NetAdapt. We also present experimental results showing quantitative comparison results of the accuracy of direct and indirect metric.
KSP Keywords
Convolution neural network(CNN), Look Up Table(LUT), Neural network design, Quantitative comparison, computational resources, mobile environment, neural network(NN)