哈佛大学提出在云、边缘与终端设备上的分布式深度神经网络

相似文件 换一批

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

1. Abstract—We propose distributed deep neural networks
   
2. (DDNNs) over distributed computing hierarchies, consisting of
   
3. the cloud, the edge (fog) and end devices. While being able to
   
4. accommodate inference of a deep neural network (DNN) in the
   
5. cloud, a DDNN also allows fast and localized inference using
   
6. shallow portions of the neural network at the edge and end
   
7. devices. When supported by a scalable distributed computing
   
8. hierarchy, a DDNN can scale up in neural network size and
   
9. scale out in geographical span. Due to its distributed nature,
   
10. DDNNs enhance sensor fusion, system fault tolerance and data
    
11. privacy for DNN applications. In implementing a DDNN, we
    
12. map sections of a DNN onto a distributed computing hierarchy.
    
13. By jointly training these sections, we minimize communication
    
14. and resource usage for devices and maximize usefulness of
    
15. extracted features which are utilized in the cloud. The resulting
    
16. system has built-in support for automatic sensor fusion and
    
17. fault tolerance. As a proof of concept, we show a DDNN
    
18. can exploit geographical diversity of sensors to improve object
    
19. recognition accuracy and reduce communication cost. In our
    
20. experiment, compared with the traditional method of offloading
    
21. raw sensor data to be processed in the cloud, DDNN locally
    
22. processes most sensor data on end devices while achieving high
    
23. accuracy and is able to reduce the communication cost by a
    
24. factor of over 20x

复制代码

本帖隐藏的内容

哈佛大学提出在云、边缘与终端设备上的分布式深度神经网络DDNN.pdf (957.7 KB)

2017-9-24 22:52:25 上传

Distributed Deep Neural Networks.pdf (1.07 MB)

2017-9-24 22:52:19 上传

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏1 回帖

谢谢楼主分享！