Edge Intelligence: Paving the Last Mile of Artificial Intelligence With Edge Computing

• ABSTRACT
  • billions of mobile and IoT devices are connected to the Internet, generating zillions bytes of data at the network edge. Driving by this trend, there is an urgent need to push the AI frontiers to the network edge
  • edge intelligence (EI)
• INTRODUCTION
  • deep learning [1]—the most dazzling sector of AI
  • As a key driver that boosts AI development, big data have recently gone through a radical shift of data source from the megascale cloud datacenters to the increasingly widespread end devices
  • However, when moving a tremendous amount of data across the wide area network (WAN), both monetary cost and transmission delay can be prohibitively high, and the privacy leakage can also be a major concern.
    • ——联邦学习不就是为了解决这个问题的么？
    • ——边缘计算和联邦学习好配啊！
    • ——再带上边缘智能？EI？edge AI？
• PRIMER ON ARTIFICIAL INTELLIGENCE（第二部分概述了人工智能的基本概念，重点是深度学习—— 人工智能中最受欢迎的领域。）
  • 深度学习让人工智能重新崛起
    The latest rise of AI after 2010s was partially due to the breakthroughs made by deep learning, a method that has achieved human-level accuracy in some interesting areas.
  • Popular Deep Learning Models
    • Convolution Neural Network
    • Recurrent Neural Network
    • Generative Adversarial Network
    • Deep Reinforcement Learning
• EDGE INTELLIGENCE（第三部分讨论了EI的动机，定义和评级。）
  The marriage of edge computing and AI gives the birth of EI
  • Specifically, edge computing aims at coordinating a multitude of collaborative edge devices and servers to process the generated data in proximity,
  • and AI strives for simulating intelligent human behavior in devices/machines by learning from data.
  • EI的六个评级（当前最高只有L3的模型架构）

    

• EDGE INTELLIGENCE MODEL TRAINING（第 IV 部分回顾了用于训练 EI 模型的体系结构、启用技术、系统和框架。）
  • Architectures
    • Centralized, Decentralized, Hybrid
  • Key Performance Indicators
    • Training Loss, Convergence, Privacy, Communication Cost, Latency, Energy Efficiency
  • Enabling Technologies
    • Federated Learning, Aggregation Frequency Control, Gradient Compression, DNN Splitting, Knowledge Transfer Learning, Gossip Training
  • Summary of Existing Systems and Frameworks
    • FedAvg, SSGD, Zoo, BlockFL, Gaia, DGC, eSGD, INCEPTIONN, Arden, PipeDream, GoSGD, Gossiping SGD, GossipGraD
    • a key challenge for distributed EI model training is the data privacy issue.
• EDGE INTELLIGENCE MODEL INFERENCE（第五部分回顾了EI模型推理的架构，启用技术，系统和框架。）
  • Edge-Based Mode, Device-Based Mode, Edge-Device Mode, Edge-Cloud Mode
  • Key Performance Indicators
    • Latency, Accuracy, Energy, Privacy, Communication Overhead, Memory Footprint
  • Enabling Technologies
    • Model Compression, Model Partition, Model Early Exit, Edge Caching, Input Filtering, Model Selection, Support for Multitenancy, Application-Specific Optimization
• FUTURE RESEARCH DIRECTIONS（第六部分讨论了EI的未来方向和挑战。）
  • Programming and Software Platforms
  • Resource-Friendly Edge AI Model Design
  • Computation-Aware Networking Techniques
  • Tradeoff Design With Various DNN Performance Metrics
  • Smart Service and Resource Management
  • Security and Privacy Issues——安全和隐私主要是以（与联邦学习有关）
  • Incentive and Business Models——激励机制与服务定价（与联邦学习有关）
• 疑问？
  • DNN划分？去查询了解！
  • top-k?
• 想法
  • 边缘网络&边缘计算&边缘智能&联邦学习
  • 深度学习用在边缘计算上=边缘智能
  • 这个边缘智能能不能和联邦学习折腾一下？
  • 数据在边缘生成，数据量又特别大，然后全部上传云端通信代价大，云端计算压力也大