Charting the Intelligence Frontiers - Edge AI Systems Nexus.

Format:

Book

Author/Creator:

Vermesan, Ovidiu.

Series:

River Publishers Series in Communications and Networking Series

Language:

English

Physical Description:

1 online resource (449 pages)

Edition:

1st ed.

Place of Publication:

Milton : River Publishers, 2026.

Summary:

This book is an essential guide for navigating, understanding, and contributing to the dynamic and rapidly evolving field of edge AI.The real value of this book lies in its innovative, forward-looking perspective, offering a guided exploration of the latest scientific breakthroughs.

Contents:

Cover

Half Title

Series Page

Title Page

Copyright Page

Dedication

Acknowledgement

Table of Contents

Preface

List of Figures

List of Tables

List of Contributors

List of Abbreviations

Chapter 1: Edge AI Systems Verification and Validation

1.1: Introduction and Background

1.2: Foundational Concepts and Edge AI Verification and Validation Taxonomy

1.2.1: Agentic AI and AI Agents

1.3: Defining Verification and Validation per Standard

1.4: Key Elements for Edge AI Verification and Validation

1.4.1: Core Elements for AI Verification

1.4.1.1: Data Verification

1.4.1.2: Model Verification

1.4.1.3: System-Level Verification

1.4.1.4: Process and Governance Verification

1.4.2: Core Elements Subject to AI Validation

1.4.2.1: Ensuring Fitness for Intended Purpose and Operational Context

1.4.2.2: Meeting User Needs and Stakeholder Expectations

1.4.2.3: Assessing Real-World Effectiveness and Outcomes

1.4.2.4: Evaluating Usability and Human-AI Interaction

1.4.2.5: Validating Ethical Alignment and Societal Impact

1.4.2.6: Data Quality and Suitability

1.5: The Edge AI Verification and Validation Lifecycle

1.6: Failure Case Behaviour in Edge-based Machine Vision Systems

1.7: Research Challenges in Edge AI Verification and Validation

1.8: Trends and Methodologies in Edge AI Verification and Validation

1.9: Conclusion

Chapter 2: Pioneering the Hybridization of Federated Learning in Human Activity Recognition

2.1: Introduction and Background

2.2: Hybrid FL Architecture

2.3: Evaluation Methodology and Metrics

2.4: Evaluation Results

2.5: Conclusion and Future Works

Chapter 3: Edge Intelligence Architecture for Distributed and Federated Learning Systems

3.1: Introduction

3.2: Related Works

3.2.1: Edge Intelligence.

3.2.2: Federated Learning

3.2.3: Model Compression

3.2.4: Beyond the State of the Art

3.3: Use Case

3.4: Architecture Proposal

3.4.1: Assumption

3.4.2: Cluster Aggregator

3.4.3: Cloud Components

3.4.4: Distributed Agent

3.5: Challenges

3.6: Conclusion

Chapter 4: Challenges and Performance of SLAM Algorithms on Resource-constrained Devices

4.1: Introduction and Background

4.2: Related Work

4.3: Methodology

4.3.1: Selected systems

4.3.2: Selected systems

4.3.3: Evaluation metrics

4.3.4: Dataset

4.4: Experimentation

4.4.1: Performance evaluation

4.4.2: TensorRT optimization for SLAM algorithms

4.5: Conclusion

4.6: Appendix

4.6.1: Calculation of the metrics used in evaluation

4.6.1.1: Absolute Trajectory Error (ATE)

4.6.1.2: Relative Pose Error (RPE)

4.6.1.3: Frames Per Second (FPS)

4.6.2: Alignment methods

4.6.2.1: Scale alignment (scale)

4.6.2.2: 6 Degress of Freedom (6DOF)

4.6.2.3: 7 Degress of Freedom (7DOF)

4.6.2.4: Scale + 7 Degrees of Freedom

Chapter 5: Designing Accelerated Edge AI Systems with Model Based Methodology

5.1: Introduction and Background

5.2: Model Based Cybertronic Systems Engineering

5.3: Designing Edge AI Systems with MBCSE Methodology

5.4: Creation of Bespoke AI Accelerator

5.4.1: High-Level Synthesis

5.4.2: Implementation and optimization with HLS

5.4.3: Verification and integration

5.5: Exemplary Results

5.6: Conclusion

Chapter 6: Edge AI Acceleration for Critical Systems: from FPGA Hardware to CGRA Technology

6.1: Introduction

6.2: State of the Art

6.3: FPG-AI: Automation Tool Flow for Efficient Deployment of Pre-trained

6.3.1: Network-in-Network (NiN) case study

6.4: GPU@SAT: RISC-V Based SoC Featuring a Soft-GPU Hardware Accelerator.

6.4.1: Enhancing a soft GPU IP reliability against SEUs in space: Modelling approach and criticality analysis on a Radiation-Tolerant FPGA

6.5: CGR-AI: Innovative Coarse-Grained Reconfigurable Array Platform

6.6: Discussion and Conclusion

Chapter 7: Model Selection and Prompting Strategies in Resource Constrained Environments for LLM-based Robotic System

7.1: Introduction and Background

7.2: Related Work

7.2.1: Local Large Language Models

7.2.2: Prompting

7.2.3: LLM in Robotics

7.3: Experimental Setup

7.3.1: System Description

7.3.2: Testing process

7.3.3: Model selection

7.3.4: Testing environment

7.4: Results

7.4.1: Differences between quantization precisions

7.4.2: Differences between models

7.4.3: Result comparison to VRAM usage

7.4.4: Result comparison to Benchmark performance

7.5: Conclusions

Chapter 8: Optimising ViT for Edge Deployment: Hybrid Token Reduction for Efficient Semantic Segmentation

8.1: Introduction and Background

8.2: Related Work

8.3: Methodology

8.3.1: Content-aware Patch Merging

8.3.2: Early-Pruning

8.4: Experiments

8.5: Conclusion

Chapter 9: Recent Trends in Edge AI: Efficient Design, Training and Deployment of Machine Learning Models

9.1: Introduction

9.2: Scalable Deep Neural Network Architectures

9.2.1: Residual networks

9.2.2: MobileNet

9.2.3: EfficientNet

9.2.4: Scalable weights

9.2.5: Practical Considerations

9.3: Neural Architecture Search for Resource Aware DNN Deployment

9.3.1: Black-Box Multi-Objective optimization

9.3.2: Differentiable NAS

9.3.3: Zero-Cost neural architecture search

9.3.4: Practical considerations

9.4: Deep Neural Network Pruning

9.4.1: Pruning granularity

9.4.2: Pruning heuristics and sensitivity analysis

9.4.3: Magnitude or threshold based heuristics.

9.4.3.1: L-Norm heuristics

9.4.3.2: Gradient Ranked Heuristics

9.4.3.3: Activation based heuristics

9.4.3.4: Relevance-based heuristics

9.4.4: Pruning schedule

9.4.5: Practical considerations

9.5: Quantization

9.5.1: Quantizers

9.5.2: Granularity

9.5.3: Methods

9.5.3.1: Post-Training quantization

9.5.3.2: Quantization-Aware training

9.5.4: Practical considerations

9.6: Cascaded Processing

9.6.1: Hierarchical systems

9.6.2: Distributed Computing

9.6.3: Early-Exit Neural Networks

9.7: Discussion

Chapter 10: Scalable Sensor Fusion for Motion Localization in Large RF Sensing Networks

10.1: Motivation

10.2: Spensor Fusion via a Probabilistic Model

10.3: Update Equations

10.3.1: Update equation for q (Cij|mi =0)

10.3.2: Update equation for q (Cij|mi =1)

10.3.3: Update equation for q (m|s =1)

10.3.4: Update equation for q (s)

10.4: Conclusions and Discussion

Chapter 11: Multi-Step Object Re-Identification on Edge Devices: A Pipeline for Vehicle Re-Identification

11.1: Introduction

11.2: Related work and state of theart

11.2.1: Object detection

11.2.2: Object feature extraction

11.2.3: Vehicle re-identification

11.2.4: Available datasets

11.2.5: Edge implementation

11.3: Proposed methodology

11.3.1: Vehicle detection, tracking and counting

11.3.2: Vehicle feature extraction and storage

11.3.2.1: Datasets

11.3.2.2: Training hyper-parameters

11.3.3: Edge device considerations

11.4: Experimental settings

11.4.1: Receiving video from a Network camera

11.4.2: Vehicle re-identification

11.4.2.1: Testing and data annotation

11.4.2.2: Saving the feature extractions

11.5: Results

11.5.1: Performance metrics

11.5.2: Dataset generalization

11.5.3: Hyper-parameters

11.5.4: Performance on the VeRi-776 benchmark.

11.5.5: Re-identification testing on test data from our cameras

11.5.5.1: Camera to camera re-identification

11.5.5.2: Sets of cameras

11.5.6: Testing the whole re-identification part of the pipeline

11.6: Future research

11.7: Conclusion

Chapter 12: A TinyMLOps Framework for Real-world Applications

12.1: Introduction

12.2: TinyMLOps methodology

12.3: A TinyMLOps framework architecture

12.4: Technology Overview for TinyMLOps Adoption

12.5: Conclusions

Chapter 13: Transfer and Self-learning in Probabilistic Models

13.1: Motivation

13.2: Prior Optimisation

13.3: Example Categorical Distribution

13.4: Conclusions and Discussion

Chapter 14: A Novel Hierarchical Approach to Perform On-device Energy Efficient Fault Classification

14.1: Introduction and Background

14.2: State of the Art

14.2.1: Experimental setup

14.2.2: Related work

14.3: Hicnn Approach

14.3.1: HiCNN training

14.3.2: Feature forwarding

14.3.3: Baseline CNN and Hierarchical CNN

14.4: Evaluation

14.4.1: Experimental setup

14.4.2: Measurement

14.5: Conclusion and Futurework

Chapter 15: Discovering and Classifying Digital and Wooden Industries Products' Defects at the Edge by a Yolo/ResNet-based Approach and Beyond

15.1: Introduction

15.2: Related Works

15.3: Spotting Defects in Wood Industry Products

15.3.1: Defect Detection Dataset

15.3.2: Experiments and Results

15.4: Spotting Defects in Digital Industry Products

15.4.1: Defect Detection and Classification Dataset

15.4.2: Experiments and Results

15.4.3: XAI Analysis: insigths into ResNet-18 using Grad-CAM

15.5: Porting of the Models on Edge Devices

15.6: Conclusions and Future Works

Chapter 16: Conscious Agents Interaction Framework for Industrial Automation

16.1: Introduction

16.2: Related Research.

16.3: Interaction Framework.

Notes:

Description based on publisher supplied metadata and other sources.

ISBN:

87-438-0888-3

87-438-0886-7

87-438-0887-5

OCLC:

1564841333