For Full-Text PDF, please login, if you are a member of IEICE,|
or go to Pay Per View on menu list, if you are a nonmember of IEICE.
Fast Inference of Binarized Convolutional Neural Networks Exploiting Max Pooling with Modified Block Structure
Ji-Hoon SHIN Tae-Hwan KIM
IEICE TRANSACTIONS on Information and Systems
Publication Date: 2020/03/01
Online ISSN: 1745-1361
Type of Manuscript: LETTER
Category: Software System
binarized neural networks, embedded systems, convolutional neural networks, inference, deep learning,
Full Text: PDF(325.6KB)>>
This letter presents a novel technique to achieve a fast inference of the binarized convolutional neural networks (BCNN). The proposed technique modifies the structure of the constituent blocks of the BCNN model so that the input elements for the max-pooling operation are binary. In this structure, if any of the input elements is +1, the result of the pooling can be produced immediately; the proposed technique eliminates such computations that are involved to obtain the remaining input elements, so as to reduce the inference time effectively. The proposed technique reduces the inference time by up to 34.11%, while maintaining the classification accuracy.