Cover
Half Title
Series Page
Title Page
Copyright Page
Dedication
Table of Contents
Preface
List of Figures
List of Tables
List of Contributors
List of Abbreviations
Chapter 1: Edge AI LoRa Mesh Technologies
	1.1: Introduction
	1.2: Overview of the State-of-the-Art Wireless Mesh Technologies
		1.2.1: Mesh components and roles
		1.2.2: Wireless routing concepts
	1.3: Routing protocols
		1.3.1: Ad hoc on-demand distance vector (AODV)
		1.3.2: Optimized link state routing (OLSR)
		1.3.3: Dynamic source routing (DSR)
		1.3.4: Routing protocol for low-power and lossy networks (RPL)
		1.3.5: Wireless mesh protocols
			1.3.5.1: B.A.T.M.A.N
			1.3.5.2: Bluetooth Low Energy
			1.3.5.3: OpenThread and Thread
			1.3.5.4: ZigBee
			1.3.5.5: Wi-Fi
			1.3.5.6: Wi-SUN
			1.3.5.7: WirelessHART
			1.3.5.8: Z-WAVE
			1.3.5.9: 6LoWPAN
	1.4: LoRa and LoRaWAN Technology
		1.4.1: LoRa physical layer
		1.4.2: LoRaWAN protocol
		1.4.3: 2.4: GHz LoRa
	1.5: LoRa Mesh and Enabling AI Technologies
	1.6: Applications for LoRa Mesh
	1.7: Conceptual Edge AI and LoRa Mesh Device Architecture
		1.7.1: Sensor and interfaces
		1.7.2: AI accelerators
		1.7.3: 2.4: GHz LoRa and Bluetooth radios
		1.7.4: Microcontrollers and microprocessors
		1.7.5: Peripheral driver
		1.7.6: Operating systems
		1.7.7: Sensor models
		1.7.8: AI learning and inference
		1.7.9: 2.4: GHz LoRa Mesh Protocol Stack
		1.7.10: AI applications and services
	1.8: Challenges and Future Research Directions
	1.9: Discussion and Conclusions
Chapter 2: Edge AI Lifecycle Management
	2.1: Introduction and Background
	2.2: Pre-development
	2.3: Development
	2.4: Production
	2.5: Conclusion
Chapter 3: Federated Learning: Privacy, Security and Hardware Perspectives
	3.1: Introduction and Background
	3.2: Federated Learning Overview
		3.2.1: Horizontal Federated Learning
		3.2.2: Vertical Federated Learning
		3.2.3: Federated Transfer Learning
	3.3: Challenges and Limitations of Federated Learning
		3.3.1: Security challenge
			3.3.1.1: Malicious Clients
			3.3.1.2: Mitigating client-based attacks
			3.3.1.3: Malicious Server attacks and mitigations
		3.3.2: Privacy challenge
			3.3.2.1: Client privacy attacks
			3.3.2.2: Mitigating client-based attacks
			3.3.2.3: Server based privacy attacks
		3.3.3: Hardware constraint and opportunities
	3.4: Conclusion
Chapter 4: Inside the AI Accelerators: From High Performance to Energy Efficiency
	4.1: Introduction and Background
	4.2: Related Work
	4.3: Classification Model
	4.4: Quantization
	4.5: Experiments and Results
		4.5.1: Time and power consumption
			4.5.1.1: Google Coral Board
			4.5.1.2: STM32MP1: Board
			4.5.1.3: NVIDIA Jetson
		4.5.2: FPGA
			4.5.2.1: QKeras Library
			4.5.2.2: Quantized model and Experimental Setup
	4.6: Conclusion
Chapter 5: Designing Lightweight CNN for Images: Architectural Components and Techniques
	5.1: Introduction and Background
	5.2: CNNs
		5.2.1: The pioneers
		5.2.2: YOLO, first step towards fast object detectors
		5.2.3: Convolutional Neural Network architecture improvements
		5.2.4: Tackling memory consumption
		5.2.5: Structural re-parameterization
	5.3: Transformers for EdgeAI
		5.3.1: Hybrid transformers
	5.4: ConvNeXts
	5.5: Neural Architecture Search
		5.5.1: NAS scale study
	5.6: Conclusion
Chapter 6: Natural Language Conditioned Planning of Complex Robotics Tasks
	6.1: Introduction
	6.2: Natural Language Processing for Robotics
		6.2.1: Large language models
		6.2.2: Multi-modal embeddings
		6.2.3: Recent implementations of high-level planning for mobile manipulation
	6.3: Action Primitives for Mobile Manipulation
		6.3.1: Methods for creating primitives
		6.3.2: Action primitive implementations
	6.4: Identified Challenges
	6.5: Conceptual Architecture
	6.6: Conclusions and Outlook
Chapter 7: An Overview of the Automated Optical Inspection Edge AI Inference System Solutions
	7.1: Introduction
	7.2: Overview of the Main Edge AI Solutions for AOI
	7.3: Comparing EdgeAI solutions for AOI
		7.3.1: Comparison using KPIs
		7.3.2: Comparison using NFRs
		7.3.3: Comparison using functional requirements
		7.3.4: Advantages of ES with respect to the other approaches
	7.4: Edge AI Solutions Demonstrator
	7.5: Conclusion
Chapter 8: Efficient AI-based Attack Detection Methods for Sensitive Edge Devices and Systems
	8.1: Introduction and Background
	8.2: Efficient Attack Detection
		8.2.1: Requirements
		8.2.2: Underlying Dataset
		8.2.3: State-of- the-Art Attack Detection Methods
		8.2.4: Selection of Applicable Algorithms
	8.3: Discussion and Conclusion
Chapter 9: Explainability and Interpretability Concepts for Edge AI Systems
	9.1: Introduction
	9.2: AI Explainability and Interpretability Goals
	9.3: AI Explainability and Interpretability Methods and Techniques
	9.4: Benchmarking
	9.5: Edge AI Explainability and Interpretability
	9.6: Challenges and Open Issues
	9.7: Conclusion
Index
About the Editors