This book provides a collection of recent research works addressing theoretical issues on improving the learning process and the generalization of GANs as well as state-of-the-art applications of GANs to various domains of real life. Adversarial learning fascinates the attention of machine learning communities across the world in recent years. Generative adversarial networks (GANs), as the main method of adversarial learning, achieve great success and popularity by exploiting a minimax learning concept, in which two networks compete with each other during the learning process. Their key capability is to generate new data and replicate available data distributions, which are needed in many practical applications, particularly in computer vision and signal processing. The book is intended for academics, practitioners, and research students in artificial intelligence looking to stay up to date with the latest advancements on GANs’ theoretical developments and their applications.
Author(s): Roozbeh Razavi-Far, Ariel Ruiz-Garcia, Vasile Palade, Juergen Schmidhuber
Series: Intelligent Systems Reference Library, 217
Publisher: Springer
Year: 2022
Language: English
Pages: 361
City: Cham
Preface
Contents
1 An Introduction to Generative Adversarial Learning: Architectures and Applications
1.1 Book Outline
References
2 Generative Adversarial Networks: A Survey on Training, Variants, and Applications
2.1 Introduction
2.2 GAN Variants
2.2.1 Original Generative Adversarial Network (GAN)
2.2.2 Deep Convolutional Generative Adversarial Network (DCGAN)
2.2.3 Conditional Generative Adversarial Network (CGAN)
2.2.4 Information Maximizing Generative Adversarial Networks (InfoGAN)
2.2.5 Auxiliary Classifier Generative Adversarial Network (ACGAN)
2.2.6 Stacked Generative Adversarial Networks (StackGAN)
2.2.7 Cycle-Consistent Generative Adversarial Network (CycleGAN)
2.2.8 Wasserstein Generative Adversarial Network (WGAN)
2.2.9 Semi-Supervised Generative Adversarial Network (SSGAN)
2.2.10 Progressive Growing Generative Adversarial Network (Progressive GAN)
2.2.11 Style-Based Generative Adversarial Network (StyleGAN)
2.2.12 Bidirectional Generative Adversarial Network (BiGAN)
2.2.13 Bayesian Generative Adversarial Network (BGAN)
2.3 GAN Applications
2.4 Conclusions
References
3 Fair Data Generation and Machine Learning Through Generative Adversarial Networks
3.1 Introduction
3.2 Overview of FairGAN Framework
3.2.1 Definitions and Metrics of Fairness
3.2.2 FairGAN Framework
3.3 Simplified Fairness Aware Generative Adversarial Networks
3.3.1 Model Framework Design
3.3.2 Empirical Evaluation
3.4 Achieving Causal Fairness in Data Generation
3.4.1 Model Framework Design
3.4.2 Achieving Direct/Indirect/Counterfactual Fairness
3.4.3 Empirical Evaluation
3.5 Achieving Fairness in Classification
3.5.1 Model Framework Design
3.5.2 Achieving Fair Classification
3.5.3 Empirical Evaluation
3.6 Related Work
3.6.1 Dealing with Different Types of Structural Data
3.6.2 Dealing with Privacy
3.7 Future Directions
3.7.1 Variants Comparison and Architecture Design
3.7.2 Achieving Long-Term Fairness in Dynamic Decision Making
3.7.3 Achieving Fairness in Regression
3.7.4 Achieving Fairness in Recommendation
3.7.5 Open Source Software
3.8 Conclusions
References
4 Quaternion Generative Adversarial Networks
4.1 Introduction
4.2 Quaternion Algebra
4.3 Generative Learning in the Quaternion Domain
4.3.1 The Quaternion Adversarial Framework
4.3.2 Quaternion Fully Connected Layers
4.3.3 Quaternion Convolutional Layers
4.3.4 Quaternion Pooling Layers
4.3.5 Quaternion Batch Normalization
4.3.6 Quaternion Spectral Normalization
4.3.7 Quaternion Weight Initialization
4.3.8 Training
4.4 GAN Architectures in the Quaternion Domain
4.4.1 Vanilla QGAN
4.4.2 Advanced QGAN
4.4.3 Evaluation Metrics
4.5 Experimental Evaluation
4.5.1 Evaluation of Spectral Normalization Methods
4.6 Conclusions
References
5 Image Generation Using Continuous Conditional Generative Adversarial Networks
5.1 Introduction and Motivation
5.2 Continuous Conditional Generative Adversarial Networks
5.2.1 Derivation of HVDL and SVDL Losses
5.2.2 A Rule of Thumb for Hyper-parameter Selection
5.2.3 Algorithms for Training CcGANs
5.3 Theoretical Analysis
5.4 Experiments
5.4.1 Case Study 1: Circular 2-D Gaussians
5.4.2 Case Study 2: UTKFace
5.5 Conclusion
References
6 Generative Adversarial Networks for Data Augmentation in Hyperspectral Image Classification
6.1 Introduction
6.1.1 Dataset
6.1.2 Data Availability
6.1.3 Class Imbalance
6.1.4 Data Augmentation
6.2 Previous Work
6.2.1 Data Augmentation
6.2.2 Class Imbalance
6.2.3 Applications in Hyperspectral Imaging
6.3 Wasserstein GAN
6.4 Conditional Wasserstein Generative Adversarial Network with Gradient Penalty for Hyperspectral Image Generation
6.4.1 Wasserstein Generative Adversarial Network with Gradient Penalty
6.4.2 Hyperspectral Data Patch
6.4.3 Dimensionality Reduction
6.4.4 Discriminator and Generator Models
6.4.5 Training Process
6.5 Experimental Results
6.5.1 Evaluation Metrics
6.5.2 Experimental Setting
6.5.3 Spectral Signature
6.5.4 Data Augmentation
6.5.5 Visualizations
6.6 Conclusion and Future Work
References
7 Face Aging Using Generative Adversarial Networks
7.1 Introduction
7.2 Generative Adversarial Networks
7.2.1 Mode Collapse
7.2.2 GANs Types
7.2.3 Reference Architectures for Facial Imaging
7.3 Databases Used for Facial Aging
7.3.1 FG-NET
7.3.2 UTKFace
7.3.3 CACD
7.3.4 MORPH
7.3.5 IMDB-Wiki
7.3.6 Cross-Age LFW (CALFW)
7.3.7 Other Databases
7.4 Experiments and Results
7.4.1 CAAE Experiments
7.4.2 IPCGAN Experiments
7.4.3 Recursive Chaining of Reversible Image-to-Image Translators Experiments, RCRIIT
7.4.4 Method Comparison
7.5 Summary
References
8 Embedding Time-Series Features into Generative Adversarial Networks for Intrusion Detection in Internet of Things Networks
8.1 Introduction
8.2 Related Works
8.2.1 Adversarially Learned Anomaly Detection
8.2.2 GAN Ensemble for Anomaly Detection
8.2.3 Anomaly Detection with GAN
8.2.4 Unsupervised Change Detection with GAN
8.2.5 Unsupervised Change Detection with GAN
8.3 Internet of Things Network
8.3.1 Dataset
8.3.2 IoT Testbed
8.4 Feature Embedding
8.4.1 Extracting Sequential Changes
8.4.2 Generative Adversarial Cluster Analysis
8.4.3 Training Computational Complexity
8.5 Experimental Results
8.5.1 Experimental Setting
8.5.2 Results Analysis
8.6 Conclusion
References
9 Inspection of Lead Frame Defects Using Deep CNN and Cycle-Consistent GAN-Based Defect Augmentation
9.1 Introduction
9.2 Defect Inspection Using the Faster R-CNN
9.2.1 Materials
9.2.2 Defect Inspection Using the Faster R-CNN
9.3 Defect Augmentation Using the CycleGAN
9.3.1 Basic GAN Structure
9.3.2 CycleGAN Structure
9.3.3 CycleGAN Training
9.3.4 CycleGAN for Image Augmentation
9.4 Experiments
9.4.1 Experimental Result of CyleGAN
9.4.2 Experimental Result of Defect Inspection Using the Faster R-CNN
9.5 Conclusions
References
10 Adversarial Learning in Accelerometer Based Transportation and Locomotion Mode Recognition
10.1 Introduction
10.2 SHL Dataset
10.3 State of the Art in Classifying the SHL Dataset
10.4 Background on Generative Adversarial Networks
10.5 Methods
10.5.1 Feature Extraction
10.5.2 Feature Scaling
10.5.3 Dealing with Class Imbalance
10.5.4 Feature Selection
10.5.5 Proposed Generative Adversarial Network
10.5.6 Hyperparameter Tuning
10.6 Implementation
10.7 Results
10.7.1 User Specific Evaluation
10.7.2 User Semi-independent Evaluation
10.7.3 User Independent Evaluation
10.8 Discussion
10.8.1 Performance Evaluation
10.8.2 Hyperparameter Tuning
10.9 Conclusion
References
11 GANs for Molecule Generation in Drug Design and Discovery
11.1 Introduction
11.2 Preliminary Concepts
11.2.1 Molecular Representation
11.2.2 Evaluation Metrics
11.3 GAN-based Molecule Generation Models
11.3.1 Goal-Directed Models
11.3.2 Distribution-Learning Models
11.3.3 Applications of GAN-based Molecule Generative Models
11.4 Comparison with Other Generative Models for Molecule Generation
11.4.1 Molecule Generation Based on Other Generative Models
11.4.2 Advantages of GAN-based Models
11.4.3 Disadvantages of GAN Based Models
11.5 Challenges and Future Directions
11.5.1 New Molecular Representations
11.5.2 New Molecule Generation Models
11.5.3 Benchmarks and Metrics
11.5.4 New Pharmaceutical Objective Functions
11.5.5 Influence of Property Prediction Models
11.6 Conclusion
References
12 Improved Diagnostic Performance of Arrhythmia Classification Using Conditional GAN Augmented Heartbeats
12.1 Introduction
12.1.1 ECG Synthesis Using Generative Adversarial Network
12.1.2 Related Work on Heartbeat Classification
12.1.3 Contributions
12.2 Data
12.2.1 Dataset Description
12.2.2 Data Preprocessing
12.3 Methodology
12.3.1 ECG Generation Methodology
12.3.2 Augmentation Using Deep Convolutional Conditional GANs
12.3.3 ECG Classification Methodology
12.4 Results and Discussion
12.4.1 Computing Platform
12.4.2 Evaluation Metrics
12.4.3 DCCGAN Performance Evaluation
12.4.4 Beat Classification Performance
12.5 Conclusion and Future Work
References
13 Generative Adversarial Network Powered Fast Magnetic Resonance Imaging—Comparative Study and New Perspectives
13.1 Introduction
13.1.1 Magnetic Resonance Imaging
13.1.2 Limitations of Magnetic Resonance Imaging
13.1.3 Conventional Acceleration Using Compressive Sensing
13.1.4 Deep Learning Based Fast MRI
13.1.5 GAN Powered Fast MRI
13.2 Methods
13.2.1 Fundamentals of MRI Reconstruction
13.2.2 CNN Based MRI Reconstruction
13.2.3 GAN Based MRI Reconstruction
13.2.4 Evaluation Methods
13.3 Benchmarking
13.4 Discussion
13.5 Conclusion
References
14 Generative Adversarial Networks for Data Augmentation in X-Ray Medical Imaging
14.1 Introduction
14.2 Previous Work on Using GANs and Transfer Learning in X-Ray Imaging
14.3 Deep Convolutional GAN (DCGAN) and Its Limitations for X-Ray Imaging
14.3.1 Architecture of DCGAN
14.3.2 DCGAN Training
14.4 Progressively Growing GAN (PGGAN)
14.4.1 Training of PGGAN
14.4.2 Phasing in a New Layer for Smooth Model Building
14.4.3 Transfer Learning for Model Building
14.5 Results and Discussion
14.5.1 Dataset
14.5.2 Augmentation Results
14.6 Conclusion
References
