Cover
Title Page
Copyright
Contents
About the Editors
List of Contributors
Preface
Part I Introduction to AI‐Based Regression and Classification
Chapter 1 Introduction to Neural Networks
1.1 Taxonomy
1.1.1 Supervised Versus Unsupervised Learning
1.1.2 Regression Versus Classification
1.1.3 Training, Validation, and Test Sets
1.2 Linear Regression
1.2.1 Objective Functions
1.2.2 Stochastic Gradient Descent
1.3 Logistic Classification
1.4 Regularization
1.5 Neural Networks
1.6 Convolutional Neural Networks
1.6.1 Convolutional Layers
1.6.2 Pooling Layers
1.6.3 Highway Connections
1.6.4 Recurrent Layers
1.7 Conclusion
References
Chapter 2 Overview of Recent Advancements in Deep Learning and Artificial Intelligence
2.1 Deep Learning
2.1.1 Supervised Learning
2.1.1.1 Conventional Approaches
2.1.1.2 Deep Learning Approaches
2.1.2 Unsupervised Learning
2.1.2.1 Algorithm
2.1.3 Toolbox
2.2 Continual Learning
2.2.1 Background and Motivation
2.2.2 Definitions
2.2.3 Algorithm
2.2.3.1 Regularization
2.2.3.2 Dynamic Network
2.2.3.3 Parameter Isolation
2.2.4 Performance Evaluation Metric
2.2.5 Toolbox
2.3 Knowledge Graph Reasoning
2.3.1 Background
2.3.2 Definitions
2.3.3 Database
2.3.4 Applications
2.3.5 Toolbox
2.4 Transfer Learning
2.4.1 Background and Motivation
2.4.2 Definitions
2.4.3 Algorithm
2.4.4 Toolbox
2.5 Physics‐Inspired Machine Learning Models
2.5.1 Background and Motivation
2.5.2 Algorithm
2.5.3 Applications
2.5.4 Toolbox
2.6 Distributed Learning
2.6.1 Introduction
2.6.2 Definitions
2.6.3 Methods
2.6.4 Toolbox
2.7 Robustness
2.7.1 Background and Motivation
2.7.2 Definitions
2.7.3 Methods
2.7.3.1 Training with Noisy Data/Labels
2.7.3.2 Adversarial Attacks
2.7.3.3 Defense Mechanisms
2.7.4 Toolbox
2.8 Interpretability
2.8.1 Background and Motivation
2.8.2 Definitions
2.8.3 Algorithm
2.8.4 ToolBox
2.9 Transformers and Attention Mechanisms for Text and Vision Models
2.9.1 Background and Motivation
2.9.2 Algorithm
2.9.3 Application
2.9.4 Toolbox
2.10 Hardware for Machine Learning Applications
2.10.1 CPU
2.10.2 GPU
2.10.3 ASICs
2.10.4 FPGA
Acknowledgment
References
Part II Advancing Electromagnetic Inverse Design with Machine Learning
Chapter 3 Breaking the Curse of Dimensionality in Electromagnetics Design Through Optimization Empowered by Machine Learning
3.1 Introduction
3.2 The SbD Pillars and Fundamental Concepts
3.3 SbD at Work in EMs Design
3.3.1 Design of Elementary Radiators
3.3.2 Design of Reflectarrays
3.3.3 Design of Metamaterial Lenses
3.3.4 Other SbD Customizations
3.4 Final Remarks and Envisaged Trends
Acknowledgments
References
Chapter 4 Artificial Neural Networks for Parametric Electromagnetic Modeling and Optimization
4.1 Introduction
4.2 ANN Structure and Training for Parametric EM Modeling
4.3 Deep Neural Network for Microwave Modeling
4.3.1 Structure of the Hybrid DNN
4.3.2 Training of the Hybrid DNN
4.3.3 Parameter‐Extraction Modeling of a Filter Using the Hybrid DNN
4.4 Knowledge‐Based Parametric Modeling for Microwave Components
4.4.1 Unified Knowledge‐Based Parametric Model Structure
4.4.2 Training with l1 Optimization of the Unified Knowledge‐Based Parametric Model
4.4.3 Automated Knowledge‐Based Model Generation
4.4.4 Knowledge‐Based Parametric Modeling of a Two‐Section Low‐Pass Elliptic Microstrip Filter
4.5 Parametric Modeling Using Combined ANN and Transfer Function
4.5.1 Neuro‐TF Modeling in Rational Form
4.5.2 Neuro‐TF Modeling in Zero/Pole Form
4.5.3 Neuro‐TF Modeling in Pole/Residue Form
4.5.4 Vector Fitting Technique for Parameter Extraction
4.5.5 Two‐Phase Training for Neuro‐TF Models
4.5.6 Neuro‐TF Model Based on Sensitivity Analysis
4.5.7 A Diplexer Example Using Neuro‐TF Model Based on Sensitivity Analysis
4.6 Surrogate Optimization of EM Design Based on ANN
4.6.1 Surrogate Optimization and Trust Region Update
4.6.2 Neural TF Optimization Method Based on Adjoint Sensitivity Analysis
4.6.3 Surrogate Model Optimization Based on Feature‐Assisted of Neuro‐TF
4.6.4 EM Optimization of a Microwave Filter Utilizing Feature‐Assisted Neuro‐TF
4.7 Conclusion
References
Chapter 5 Advanced Neural Networks for Electromagnetic Modeling and Design
5.1 Introduction
5.2 Semi‐Supervised Neural Networks for Microwave Passive Component Modeling
5.2.1 Semi‐Supervised Learning Based on Dynamic Adjustment Kernel Extreme Learning Machine
5.2.1.1 Dynamic Adjustment Kernel Extreme Learning Machine
5.2.1.2 Semi‐Supervised Learning Based on DA‐KELM
5.2.1.3 Numerical Examples
5.2.2 Semi‐Supervised Radial Basis Function Neural Network
5.2.2.1 Semi‐Supervised Radial Basis Function Neural Network
5.2.2.2 Sampling Strategy
5.2.2.3 SS‐RBFNN With Sampling Strategy
5.3 Neural Networks for Antenna and Array Modeling
5.3.1 Modeling of Multiple Performance Parameters for Antennas
5.3.2 Inverse Artificial Neural Network for Multi‐objective Antenna Design
5.3.2.1 Knowledge‐Based Neural Network for Periodic Array Modeling
5.4 Autoencoder Neural Network for Wave Propagation in Uncertain Media
5.4.1 Two‐Dimensional GPR System with the Dispersive and Lossy Soil
5.4.2 Surrogate Model for GPR Modeling
5.4.3 Modeling Results
References
Part III Deep Learning for Metasurface Design
Chapter 6 Generative Machine Learning for Photonic Design
6.1 Brief Introduction to Generative Models
6.1.1 Probabilistic Generative Model
6.1.2 Parametrization and Optimization with Generative Models
6.1.2.1 Probabilistic Model for Gradient‐Based Optimization
6.1.2.2 Sampling‐Based Optimization
6.1.2.3 Generative Design Strategy
6.1.2.4 Generative Adversarial Networks in Photonic Design
6.1.2.5 Discussion
6.2 Generative Model for Inverse Design of Metasurfaces
6.2.1 Generative Design Strategy for Metasurfaces
6.2.2 Model Validation
6.2.3 On‐demand Design Results
6.3 Gradient‐Free Optimization with Generative Model
6.3.1 Gradient‐Free Optimization Algorithms
6.3.2 Evolution Strategy with Generative Parametrization
6.3.2.1 Generator from VAE
6.3.2.2 Evolution Strategy
6.3.2.3 Model Validation
6.3.2.4 On‐demand Design Results
6.3.3 Cooperative Coevolution and Generative Parametrization
6.3.3.1 Cooperative Coevolution
6.3.3.2 Diatomic Polarizer
6.3.3.3 Gradient Metasurface
6.4 Design Large‐Scale, Weakly Coupled System
6.4.1 Weak Coupling Approximation
6.4.2 Analog Differentiator
6.4.3 Multiplexed Hologram
6.5 Auxiliary Methods for Generative Photonic Parametrization
6.5.1 Level Set Method
6.5.2 Fourier Level Set
6.5.3 Implicit Neural Representation
6.5.4 Periodic Boundary Conditions
6.6 Summary
References
Chapter 7 Machine Learning Advances in Computational Electromagnetics
7.1 Introduction
7.2 Conventional Electromagnetic Simulation Techniques
7.2.1 Finite Difference Frequency (FDFD) and Time (FDTD) Domain Solvers
7.2.2 The Finite Element Method (FEM)
7.2.2.1 Meshing
7.2.2.2 Basis Function Expansion
7.2.2.3 Residual Formulation
7.2.3 Method of Moments (MoM)
7.3 Deep Learning Methods for Augmenting Electromagnetic Solvers
7.3.1 Time Domain Simulators
7.3.1.1 Hardware Acceleration
7.3.1.2 Learning Finite Difference Kernels
7.3.1.3 Learning Absorbing Boundary Conditions
7.3.2 Augmenting Variational CEM Techniques Via Deep Learning
7.4 Deep Electromagnetic Surrogate Solvers Trained Purely with Data
7.5 Deep Surrogate Solvers Trained with Physical Regularization
7.5.1 Physics‐Informed Neural Networks (PINNs)
7.5.2 Physics‐Informed Neural Networks with Hard Constraints (hPINNs)
7.5.3 WaveY‐Net
7.6 Conclusions and Perspectives
Acknowledgments
References
Chapter 8 Design of Nanofabrication‐Robust Metasurfaces Through Deep Learning‐Augmented Multiobjective Optimization
8.1 Introduction
8.1.1 Metasurfaces
8.1.2 Fabrication State‐of‐the‐Art
8.1.3 Fabrication Challenges
8.1.3.1 Fabrication Defects
8.1.4 Overcoming Fabrication Limitations
8.2 Related Work
8.2.1 Robustness Topology Optimization
8.2.2 Deep Learning in Nanophotonics
8.3 DL‐Augmented Multiobjective Robustness Optimization
8.3.1 Supercells
8.3.1.1 Parameterization of Freeform Meta‐Atoms
8.3.2 Robustness Estimation Method
8.3.2.1 Simulating Defects
8.3.2.2 Existing Estimation Methods
8.3.2.3 Limitations of Existing Methods
8.3.2.4 Solver Choice
8.3.3 Deep Learning Augmentation
8.3.3.1 Challenges
8.3.3.2 Method
8.3.4 Multiobjective Global Optimization
8.3.4.1 Single Objective Cost Functions
8.3.4.2 Dominance Relationships
8.3.4.3 A Robustness Objective
8.3.4.4 Problems with Optimization and DL Models
8.3.4.5 Error‐Tolerant Cost Functions
8.3.5 Robust Supercell Optimization
8.3.5.1 Pareto Front Results
8.3.5.2 Examples from the Pareto Front
8.3.5.3 The Value of Exhaustive Sampling
8.3.5.4 Speedup Analysis
8.4 Conclusion
8.4.1 Future Directions
Acknowledgments
References
Chapter 9 Machine Learning for Metasurfaces Design and Their Applications
9.1 Introduction
9.1.1 ML/DL for RIS Design
9.1.2 ML/DL for RIS Applications
9.1.3 Organization
9.2 Inverse RIS Design
9.2.1 Genetic Algorithm (GA)
9.2.2 Particle Swarm Optimization (PSO)
9.2.3 Ant Colony Optimization (ACO)
9.3 DL‐Based Inverse Design and Optimization
9.3.1 Artificial Neural Network (ANN)
9.3.1.1 Deep Neural Networks (DNN)
9.3.2 Convolutional Neural Networks (CNNs)
9.3.3 Deep Generative Models (DGMs)
9.3.3.1 Generative Adversarial Networks (GANs)
9.3.3.2 Conditional Variational Autoencoder (cVAE)
9.3.3.3 Global Topology Optimization Networks (GLOnets)
9.4 Case Studies
9.4.1 MTS Characterization Model
9.4.2 Training and Design
9.5 Applications
9.5.1 DL‐Based Signal Detection in RIS
9.5.2 DL‐Based RIS Channel Estimation
9.6 DL‐Aided Beamforming for RIS Applications
9.6.1 Beamforming at the RIS
9.6.2 Secure‐Beamforming
9.6.3 Energy‐Efficient Beamforming
9.6.4 Beamforming for Indoor RIS
9.7 Challenges and Future Outlook
9.7.1 Design
9.7.1.1 Hybrid Physics‐Based Models
9.7.1.2 Other Learning Techniques
9.7.1.3 Improved Data Representation
9.7.2 Applications
9.7.3 Channel Modeling
9.7.3.1 Data Collection
9.7.3.2 Model Training
9.7.3.3 Environment Adaptation and Robustness
9.8 Summary
Acknowledgments
References
Part IV RF, Antenna, Inverse‐Scattering, and Other EM Applications of Deep Learning
Chapter 10 Deep Learning for Metasurfaces and Metasurfaces for Deep Learning
10.1 Introduction
10.2 Forward‐Predicting Networks
10.2.1 FCNN (Fully Connected Neural Networks)
10.2.2 CNN (Convolutional Neural Networks)
10.2.2.1 Nearly Free‐Form Meta‐Atoms
10.2.2.2 Mutual Coupling Prediction
10.2.3 Sequential Neural Networks and Universal Forward Prediction
10.2.3.1 Sequencing Input Data
10.2.3.2 Recurrent Neural Networks
10.2.3.3 1D Convolutional Neural Networks
10.3 Inverse‐Design Networks
10.3.1 Tandem Network for Inverse Designs
10.3.2 Generative Adversarial Nets (GANs)
10.4 Neuromorphic Photonics
10.5 Summary and Outlook
References
Chapter 11 Forward and Inverse Design of Artificial Electromagnetic Materials
11.1 Introduction
11.1.1 Problem Setting
11.1.2 Artificial Electromagnetic Materials
11.1.2.1 Regime 1: Floquet–Bloch
11.1.2.2 Regime 2: Resonant Effective Media
11.1.2.3 All‐Dielectric Metamaterials
11.2 The Design Problem Formulation
11.3 Forward Design
11.3.1 Search Efficiency
11.3.2 Evaluation Time
11.3.3 Challenges with the Forward Design of Advanced AEMs
11.3.4 Deep Learning the Forward Model
11.3.4.1 When Does Deep Learning Make Sense?
11.3.4.2 Common Deep Learning Architectures
11.3.5 The Forward Design Bottleneck
11.4 Inverse Design with Deep Learning
11.4.1 Why Inverse Problems Are Often Difficult
11.4.2 Deep Inverse Models
11.4.2.1 Does the Inverse Model Address Non‐uniqueness?
11.4.2.2 Multi‐solution Versus Single‐Solution Models
11.4.2.3 Iterative Methods versus Direct Mappings
11.4.3 Which Inverse Models Perform Best?
11.5 Conclusions and Perspectives
11.5.1 Reducing the Need for Training Data
11.5.1.1 Transfer Learning
11.5.1.2 Active Learning
11.5.1.3 Physics‐Informed Learning
11.5.2 Inverse Modeling for Non‐existent Solutions
11.5.3 Benchmarking, Replication, and Sharing Resources
Acknowledgments
References
Chapter 12 Machine Learning‐Assisted Optimization and Its Application to Antenna and Array Designs
12.1 Introduction
12.2 Machine Learning‐Assisted Optimization Framework
12.3 Machine Learning‐Assisted Optimization for Antenna and Array Designs
12.3.1 Design Space Reduction
12.3.2 Variable‐Fidelity Evaluation
12.3.3 Hybrid Optimization Algorithm
12.3.4 Robust Design
12.3.5 Antenna Array Synthesis
12.4 Conclusion
References
Chapter 13 Analysis of Uniform and Non‐uniform Antenna Arrays Using Kernel Methods
13.1 Introduction
13.2 Antenna Array Processing
13.2.1 Detection of Angle of Arrival
13.2.2 Optimum Linear Beamformers
13.2.3 Direction of Arrival Detection with Random Arrays
13.3 Support Vector Machines in the Complex Plane
13.3.1 The Support Vector Criterion for Robust Regression in the Complex Plane
13.3.2 The Mercer Theorem and the Nonlinear SVM
13.4 Support Vector Antenna Array Processing with Uniform Arrays
13.4.1 Kernel Array Processors with Temporal Reference
13.4.1.1 Relationship with the Wiener Filter
13.4.2 Kernel Array Processor with Spatial Reference
13.4.2.1 Eigenanalysis in a Hilbert Space
13.4.2.2 Formulation of the Processor
13.4.2.3 Relationship with Nonlinear MVDM
13.4.3 Examples of Temporal and Spatial Kernel Beamforming
13.5 DOA in Random Arrays with Complex Gaussian Processes
13.5.1 Snapshot Interpolation from Complex Gaussian Process
13.5.2 Examples
13.6 Conclusion
Acknowledgments
References
Chapter 14 Knowledge‐Based Globalized Optimization of High‐Frequency Structures Using Inverse Surrogates
14.1 Introduction
14.2 Globalized Optimization by Feature‐Based Inverse Surrogates
14.2.1 Design Task Formulation
14.2.2 Evaluating Design Quality with Response Features
14.2.3 Globalized Search by Means of Inverse Regression Surrogates
14.2.4 Local Tuning Procedure
14.2.5 Global Optimization Algorithm
14.3 Results
14.3.1 Verification Structures
14.3.2 Results
14.3.3 Discussion
14.4 Conclusion
Acknowledgment
References
Chapter 15 Deep Learning for High Contrast Inverse Scattering of Electrically Large Structures
15.1 Introduction
15.2 General Strategy and Approach
15.2.1 Related Works by Others and Corresponding Analyses
15.2.2 Motivation
15.3 Our Approach for High Contrast Inverse Scattering of Electrically Large Structures
15.3.1 The 2‐D Inverse Scattering Problem with Electrically Large Structures
15.3.1.1 Dual‐Module NMM‐IEM Machine Learning Model
15.3.1.2 Receiver Approximation Machine Learning Method
15.3.2 Application for 3‐D Inverse Scattering Problem with Electrically Large Structures
15.3.2.1 Semi‐Join Extreme Learning Machine
15.3.2.2 Hybrid Neural Network Electromagnetic Inversion Scheme
15.4 Applications of Our Approach
15.4.1 Applications for 2‐D Inverse Scattering Problem with Electrically Large Structures
15.4.1.1 Dual‐Module NMM‐IEM Machine Learning for Fast Electromagnetic Inversion of Inhomogeneous Scatterers with High Contrasts and Large Electrical Dimensions
15.4.1.2 Nonlinear Electromagnetic Inversion of Damaged Experimental Data by a Receiver Approximation Machine Learning Method
15.4.2 Applications for 3‐D Inverse Scattering Problem with Electrically Large Structures
15.4.2.1 Super‐Resolution 3‐D Microwave Imaging of Objects with High Contrasts by a Semi‐Join Extreme Learning Machine
15.4.2.2 A Hybrid Neural Network Electromagnetic Inversion Scheme (HNNEMIS) for Super‐Resolution 3‐Dimensional Microwave Human Brain Imaging
15.5 Conclusion and Future work
15.5.1 Summary of Our Work
15.5.1.1 Limitations and Potential Future Works
References
Chapter 16 Radar Target Classification Using Deep Learning
16.1 Introduction
16.2 Micro‐Doppler Signature Classification
16.2.1 Human Motion Classification
16.2.2 Human Hand Gesture Classification
16.2.3 Drone Detection
16.3 SAR Image Classification
16.3.1 Vehicle Detection
16.3.2 Ship Detection
16.4 Target Classification in Automotive Radar
16.5 Advanced Deep Learning Algorithms for Radar Target Classification
16.5.1 Transfer Learning
16.5.2 Generative Adversarial Networks
16.5.3 Continual Learning
16.6 Conclusion
References
Chapter 17 Koopman Autoencoders for Reduced‐Order Modeling of Kinetic Plasmas
17.1 Introduction
17.2 Kinetic Plasma Models: Overview
17.3 EMPIC Algorithm
17.3.1 Overview
17.3.2 Field Update Stage
17.3.3 Field Gather Stage
17.3.4 Particle Pusher Stage
17.3.5 Current and Charge Scatter Stage
17.3.6 Computational Challenges
17.4 Koopman Autoencoders Applied to EMPIC Simulations
17.4.1 Overview and Motivation
17.4.2 Koopman Operator Theory
17.4.3 Koopman Autoencoder (KAE)
17.4.3.1 Case Study I: Oscillating Electron Beam
17.4.3.2 Case Study II: Virtual Cathode Formation
17.4.4 Computational Gain
17.5 Towards A Physics‐Informed Approach
17.6 Outlook
Acknowledgments
References
Index
EULA