Note: Intro -- Preface -- Acknowledgements -- Contents -- Acronyms -- Part I Introduction and Fundamentals -- 1 Introduction -- 1.1 A Brief Introduction to Edge Computing -- 1.2 Trends in Edge Computing -- 1.3 Industrial Applications of Edge Computing -- 1.4 Intelligent Edge and Edge Intelligence -- References -- 2 Fundamentals of Edge Computing -- 2.1 Paradigms of Edge Computing -- 2.1.1 Cloudlet and Micro Data Centers -- 2.1.2 Fog Computing -- 2.1.3 Mobile and Multi-Access Edge Computing (MEC) -- 2.1.4 Definition of Edge Computing Terminologies -- 2.1.5 Collaborative End-Edge-Cloud Computing -- 2.2 Hardware for Edge Computing -- 2.2.1 AI Hardware for Edge Computing -- 2.2.2 Integrated Commodities Potentially for Edge Nodes -- 2.3 Edge Computing Frameworks -- 2.4 Virtualizing the Edge -- 2.4.1 Virtualization Techniques -- 2.4.2 Network Virtualization -- 2.4.3 Network Slicing -- 2.5 Value Scenarios for Edge Computing -- 2.5.1 Smart Parks -- 2.5.2 Video Surveillance -- 2.5.3 Industrial Internet of Things -- References -- 3 Fundamentals of Artificial Intelligence -- 3.1 Artificial Intelligence and Deep Learning -- 3.2 Neural Networks in Deep Learning -- 3.2.1 Fully Connected Neural Network (FCNN) -- 3.2.2 Auto-Encoder (AE) -- 3.2.3 Convolutional Neural Network (CNN) -- 3.2.4 Generative Adversarial Network (GAN) -- 3.2.5 Recurrent Neural Network (RNN) -- 3.2.6 Transfer Learning (TL) -- 3.3 Deep Reinforcement Learning (DRL) -- 3.3.1 Reinforcement Learning (RL) -- 3.3.2 Value-Based DRL -- 3.3.3 Policy-Gradient-Based DRL -- 3.4 Distributed DL Training -- 3.4.1 Data Parallelism -- 3.4.2 Model Parallelism -- 3.5 Potential DL Libraries for Edge -- References -- Part II Artificial Intelligence and Edge Computing -- 4 Artificial Intelligence Applications on Edge -- 4.1 Real-time Video Analytic -- 4.1.1 Machine Learning Solution -- 4.1.2 Deep Learning Solution. , 4.1.2.1 End Level -- 4.1.2.2 Edge Level -- 4.1.2.3 Cloud Level -- 4.2 Autonomous Internet of Vehicles (IoVs) -- 4.2.1 Machine Learning Solution -- 4.2.2 Deep Learning Solution -- 4.2.2.1 End Level -- 4.2.2.2 Edge Level -- 4.2.2.3 Cloud Level -- 4.3 Intelligent Manufacturing -- 4.3.1 Machine Learning Solution -- 4.3.2 Deep Learning Solution -- 4.3.2.1 End Level -- 4.3.2.2 Edge Level -- 4.3.2.3 Cloud Level -- 4.4 Smart Home and City -- 4.4.1 Machine Learning Solution -- 4.4.2 Deep Learning Solution -- 4.4.2.1 End Level -- 4.4.2.2 Edge Level -- 4.4.2.3 Cloud Level -- References -- 5 Artificial Intelligence Inference in Edge -- 5.1 Optimization of AI Models in Edge -- 5.1.1 General Methods for Model Optimization -- 5.1.2 Model Optimization for Edge Devices -- 5.2 Segmentation of AI Models -- 5.3 Early Exit of Inference (EEoI) -- 5.4 Sharing of AI Computation -- References -- 6 Artificial Intelligence Training at Edge -- 6.1 Distributed Training at Edge -- 6.2 Vanilla Federated Learning at Edge -- 6.3 Communication-Efficient FL -- 6.4 Resource-Optimized FL -- 6.5 Security-Enhanced FL -- 6.6 A Case Study for Training DRL at Edge -- 6.6.1 Multi-User Edge Computing Scenario -- 6.6.2 System Formulation -- 6.6.3 Offloading Strategy for Computing Tasks Based on DRL -- 6.6.4 Distributed Cooperative Training -- References -- 7 Edge Computing for Artificial Intelligence -- 7.1 Edge Hardware for AI -- 7.1.1 Mobile CPUs and GPUs -- 7.1.2 FPGA-Based Solutions -- 7.1.3 TPU-Based Solutions -- 7.2 Edge Data Analysis for Edge AI -- 7.2.1 Challenge and Needs for Edge Data Process -- 7.2.2 Combination of Big Data and Edge Data Process -- 7.2.3 Architecture for Edge Data Process -- 7.3 Communication and Computation Modes for Edge AI -- 7.3.1 Integral Offloading -- 7.3.2 Partial Offloading -- 7.3.3 Vertical Collaboration -- 7.3.4 Horizontal Collaboration. , 7.4 Tailoring Edge Frameworks for AI -- 7.5 Performance Evaluation for Edge AI -- References -- 8 Artificial Intelligence for Optimizing Edge -- 8.1 AI for Adaptive Edge Caching -- 8.1.1 Use Cases of DNNs -- 8.1.2 Use Cases of DRL -- 8.2 AI for Optimizing Edge Task Offloading -- 8.2.1 Use Cases of DNNs -- 8.2.2 Use Cases of DRL -- 8.3 AI for Edge Management and Maintenance -- 8.3.1 Edge Communication -- 8.3.2 Edge Security -- 8.3.3 Joint Edge Optimization -- 8.4 A Practical Case for Adaptive Edge Caching -- 8.4.1 Multi-BS Edge Caching Scenario -- 8.4.2 System Formulation -- 8.4.3 Weighted Distributed DQN Training and Cache Replacement -- 8.4.4 Conclusion for Edge Caching Case -- References -- Part III Challenges and Conclusions -- 9 Lessons Learned and Open Challenges -- 9.1 More Promising Applications -- 9.2 General AI Model for Inference -- 9.2.1 Ambiguous Performance Metrics -- 9.2.2 Generalization of EEoI -- 9.2.3 Hybrid Model Modification -- 9.2.4 Coordination Between AI Training and Inference -- 9.3 Complete Edge Architecture for AI -- 9.3.1 Edge for Data Processing -- 9.3.2 Microservice for Edge AI Services -- 9.3.3 Incentive and Trusty Offloading Mechanism for AI -- 9.3.4 Integration with ``AI for Optimizing Edge'' -- 9.4 Practical Training Principles at Edge -- 9.4.1 Data Parallelism Versus Model Parallelism -- 9.4.2 Training Data Resources -- 9.4.3 Asynchronous FL at Edge -- 9.4.4 Transfer Learning-Based Training -- 9.5 Deployment and Improvement of Intelligent Edge -- References -- 10 Conclusions.