Scholarlink

In this chapter, we focus on distributed training of DNN at the edge, including the architectures, key performance indicators, enabling techniques and existing systems & frameworks. We clasify the architectures of distributed DNN training into three categories: Centralized, Decentralized, Hybrid (Cloud-Edge-Device), as illustrated by subfigures (a), (b) and (c) in Fig. 2.1, respectively. The cloud refers to the central datacenter whereas the end devices are represented by mobile phones, cars and surveillance cameras, which are also data sources. For the edge server, we use base stations as the legend. As shown in Fig. 2.1, all three architectures involve cooperation between different kinds of devices. However, only the devices with a neural network mark process DNN training, either partially or wholly.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Edge Intelligence via Model Training

  • 章节

  • 发表日期：

    2025年

查看全文
  • Sen Lin,
  • Zhi Zhou,
  • Zhaofeng Zhang,
  • Xu Chen,
  • Junshan Zhang

摘要

In this chapter, we focus on distributed training of DNN at the edge, including the architectures, key performance indicators, enabling techniques and existing systems & frameworks. We clasify the architectures of distributed DNN training into three categories: Centralized, Decentralized, Hybrid (Cloud-Edge-Device), as illustrated by subfigures (a), (b) and (c) in Fig. 2.1, respectively. The cloud refers to the central datacenter whereas the end devices are represented by mobile phones, cars and surveillance cameras, which are also data sources. For the edge server, we use base stations as the legend. As shown in Fig. 2.1, all three architectures involve cooperation between different kinds of devices. However, only the devices with a neural network mark process DNN training, either partially or wholly.