[PART 1 GANs and Generative Modeling]

CHAPTER 1: GETTING STARTED WITH GAN
__1.1 What is GAN?
__1.2 How GANs Work
__1.3 GAN system
__1.4 Why should we study GANs?
__1.5 In conclusion

CHAPTER 2 Autoencoders and Generative Learning
__2.1 Getting Started with Generative Modeling
__2.2 How an autoencoder works
__2.3 Comparing GANs and Autoencoders
__2.4 Autoencoder configuration
__2.5 Using autoencoders
__2.6 Unsupervised Learning
__2.7 Code is the key
__2.8 Why GAN?
__2.9 In conclusion

CHAPTER 3: Implementing Your First GAN
__3.1 GAN Basics: Adversarial Training
__3.2 Constructors and Discriminators
__3.3 GAN training algorithm
__3.4 Tutorial: Generating Handwritten Numbers
__3.5 Conclusion
__3.6 In conclusion

CHAPTER 4 DCGAN
__4.1 Convolutional Neural Networks
__4.2 A Brief History of DCGAN
__4.3 Batch Normalization
__4.4 Tutorial: Generating Handwritten Digits with DCGAN
__4.5 Conclusion
__4.6 In conclusion


[PART 2: The Latest GAN Model]

CHAPTER 5: The Challenges and Know-How of GAN Training
__5.1 rating
__5.2 Difficulties in Training
__5.3 Game Settings Summary
__5.4 Training Know-How
__5.5 In conclusion

CHAPTER 6 ProGAN
__6.1 Latent space interpolation
__6.2 Amazing speed of development
__6.3 Summary of Key Innovations
__6.4 Hands-on with TensorFlow Hub
__6.5 Practical Applications
__6.6 In conclusion

CHAPTER 7 SGAN
__7.1 Introduction to SGAN
__7.2 Tutorial: Implementing SGAN
__7.3 Comparison with supervised learning classifiers
__7.4 Conclusion
__7.5 In conclusion

CHAPTER 8 CGAN
__8.1 Motivation
__8.2 Introduction to CGAN
__8.3 Tutorial: Implementing CGAN
__8.4 Conclusion
__8.5 In conclusion

CHAPTER 9 CycleGAN
__9.1 Image-to-Image Conversion
__9.2 Cycle-Consistency Loss: Going and Coming Back
__9.3 Hostile losses
__9.4 Loss of identity
__9.5 Structure
__9.6 Implementing GAN in an Object-Oriented Way
__9.7 Tutorial: Implementing CycleGAN
__9.8 CycleGAN's Extensions and Applications
__9.9 In conclusion


[PART 3: What You Will Learn Next]

CHAPTER 10 Hostile Samples
__10.1 Introducing the Hostile Sample
__10.2 Predictions, Bad Predictions, Distributions
__10.3 Right Training and Wrong Training
__10.4 Signal and Noise
__10.5 New Hope
__10.6 From Adversarial Samples to GANs
__10.7 Conclusion
__10.8 In conclusion

CHAPTER 11 Practical GAN ​​Applications
__11.1 GANs in the Medical Field
__11.2 GANs in Fashion
__11.3 Conclusion
__11.4 In conclusion

CHAPTER 12: FUTURE PROSPECTS
__12.1 Ethics
__12.2 Three Innovations in GAN
__12.3 Further Reading
__12.4 Summary
__12.5 In conclusion

Key Contents
_ Understanding how GANs work and their generators and discriminators
_ Generating Handwritten Digits with Autoencoders and GANs
_ Implementing GANs with CNNs and DCGANs and Understanding Batch Normalization
_ Generating high-resolution images with ProGAN
_ Understanding SGANs in Semi-Supervised Learning
_ Create the desired handwritten number image using CGAN
_ Turning apples into oranges and oranges into apples with CycleGAN
_ Understanding the challenges of GAN training and generating adversarial samples with real images and noise
_ Examining the application and use cases of GANs in the medical and fashion fields.


A next-level neural network that learns and evolves on its own
GANs are a type of self-critical machine learning system.
This is something I've always been disappointed with, as it's not found in other machine learning methods.
People constantly make possible plans and determine whether they are feasible or not.
And I know very well that just jumping into work blindly isn't the answer.
In that sense, GANs are truly a logical neural network for implementing a higher level of artificial intelligence, as they can leverage automatically learned representations and machine learning feedback loops.


There's not much new in other parts of machine learning anymore.
Most of the concepts in the field of computer vision were conceived before 1998.
On the other hand, what GANs do was impossible before 2014. Since their inception, GANs have continued to grow exponentially, even as I write this.

GANs are an exciting new world with many possibilities.
I am honored and excited to share this with you.
It took me nearly two years to write this book.
We hope you enjoy reading this book as much as we did.
I can't wait to see the amazing inventions you'll bring to the world in the future.

- From the author's note


Target audience
This book is intended for those with experience with machine learning and neural networks.
I've done my best to cover what's needed in each chapter of the book, but you should definitely know at least 70% of what's listed below.

1.
Ability to create intermediate or higher level Python programs
2.
Understanding object-oriented programming, how to handle objects, properties, and methods
3.
Machine learning fundamentals, including training/test dataset separation, overfitting, weights, and hyperparameters.
4.
Basic statistics and calculus, such as probability, density functions, probability distributions, differentiation, and simple optimization.
5.
A foundation in linear algebra, such as matrices, high-dimensional spaces, and (ideally) principal component analysis.
6.
Deep learning fundamentals, including feedforward neural networks, weights and biases, activation functions, regularization, and stochastic gradient descent.
7.
Experience using Keras at least a little or willingness to learn it separately


Structure of this book
This book covers theory and practice in a balanced way and is divided into three parts.

Part 1: GANs and Generative Modeling
We will look at the basic concepts of generative learning and GANs and implement the most basic GAN model.


Chapter 1: Getting Started with GANs
Introduces GANs and explains how they work at a high level.
We will explore how generator and discriminator networks compete to train.


Chapter 2 Autoencoders and Generative Learning
Let's first look at the autoencoder, which can be considered the pioneer of GAN.
Let's generate handwritten numbers using variational autoencoder (VAE).

_Chapter 3: Implementing Your First GAN
This course delves into the theory behind GANs and adversarial learning. It explores the key differences between GANs and traditional neural networks, and explores the differences in their cost functions and training processes.
Let's implement GAN with Keras and generate handwritten numbers.


Chapter 4 DCGAN
Introducing convolutional neural networks (CNNs) and batch normalization.
Next, we implement DCGAN, an advanced GAN structure that utilizes batch normalization to stabilize the training process.

Part 2: The Latest GAN Model
Building on the basics learned in Part 1, we will delve deeper into GAN theory and implement advanced GAN structures.

Chapter 5: The Difficulties and Know-How of GAN Training
We will explore the theoretical and practical challenges encountered in training GANs and how to overcome them.
We will explore best practices based on academic papers and presentations, and also cover how to measure the performance of GANs.


Chapter 6 ProGAN
We explore ProGAN, a state-of-the-art method for training generators and discriminators.
ProGAN adds new layers during the training process to generate images with superior quality and resolution.

Chapter 7 SGAN
We'll learn how to improve classification accuracy with a small amount of labeled training data using semi-supervised learning. We'll implement SGANs and explore how to leverage labels to turn the discriminator into a powerful multi-class classifier.

Chapter 8 CGAN
CGANs overcome the drawback of generative modeling, which cannot precisely specify which samples to synthesize using labels or other conditional information, by training the generator and discriminator. We will examine the process of implementing CGANs to directly generate desired data.

Chapter 9 CycleGAN
We explore the innovation of CycleGAN, which can be used to transform an image into another image.
This includes changing a picture of a horse into a picture of a zebra, or changing an apple into an orange, or an orange into an apple.

Part 3: What You Will Learn Next
We will examine how to utilize GANs and adversarial learning and examples of their use.


Chapter 10 Hostile Sample
We'll look at adversarial samples, a technique for intentionally tricking machine learning models into making mistakes.
We explain the importance of adversarial samples with theoretical and practical examples and explore their relevance to GANs.

Chapter 11: Practical GAN ​​Applications
We'll explore best practices for how the technologies discussed above are applied in the healthcare and fashion industries.
In the medical field, we will look at how GANs can be used to augment small amounts of data, and in the fashion field, we will look at how they can be used to personalize content.

Chapter 12: Future Prospects
We conclude by summarizing the main points of the book and discussing the ethical aspects of GANs.
For those who want to continue exploring this field, we also introduce emerging GAN techniques.


Related Books (Title + ISBN)

Hands-On Machine Learning (2nd Edition) / 9791162242964
GAN Deep Learning Practical Project at an Art Museum / 9791162241080
Deep Learning from Scratch 2 / 9791162241745