Compacting Deep Neural Networks for Light Weight IoT & SCADA Based Applications with Node Pruning

Abstract

Deeplearning based image classifier is getting improved day by day. The network architecture is also increasing with the accuracy. But the bigger size and resource intensive training makes this model impractical to deploy in IoT based computational units. IoT has limited resources and reckoning power. So smaller network with same accuracy is highly priced for IoT based application deployment. In this study, convolutional deeplearning neural network and how pruning filters without compromising accuracy was studied. Efficient result was achieved from the pruned deeplearning neural network. the model was configured in the experiments by pruning the filter based on absolution position of zeros value based filter ranking. SCADA applications with intelligent component to detect data abnormality and remote sensing also required neural network applications. Using compact memory efficient module in such machines will also give proper validation in such applications in real time. In the end, proposed method for the pruned network delivered same accuracy with reduced size and thus archiving memory and computation for small sized application.