首页> 外文会议>電子情報通信学会;情報科学技術フォーラム >A Comprehensive Analysis of Low-impact Computations in Deep Learning Applications
【24h】

A Comprehensive Analysis of Low-impact Computations in Deep Learning Applications

机译:深度学习应用中低影响计算的全面分析

获取原文

摘要

Analyses of rate of Zero value and time consumptionreveal redundancy covolutional computation and are assumedcomputational volume reduction of CNN. Average rates ofZero value in convolution layers are 35% and 55%, For fullyconnection layers, the average rates are 47% and 33%. it isproved by redundancy computation. Largest time consumptionis conv1 layer, and this time is about 3 times larger than secondone for the reason of the total number of multiplications.By reducing the redundancy, with faster execution speed, itcan be led performance efficiency of CNN can be improved.
机译:分析零值和时间消耗率揭示冗余覆盖计算并被假设CNN的计算量减少。平均价格卷积层中的零值为35%和55%,适用于完全连接层,平均速率为47%和33%。这是通过冗余计算证明。最大的消耗量是Conv1层,这个时间大约是秒大约3倍一个是出于乘法总数的原因。通过减少冗余,执行速度更快可以是CNN的LED性能效率可以提高。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号