Generative adversarial networks (GANs) are a type of neural network introduced in 2014. GANs consist of two neural networks, a generator and discriminator, that compete against each other. The generator creates new data instances to fool the discriminator, while the discriminator evaluates whether instances are real or generated. Through this adversarial training process, GANs can generate highly realistic new images, text, and other data types. Common applications of GANs include image-to-image translation, super resolution, text-to-image generation, and more. Researchers continue advancing GAN techniques and exploring new applications.

1. GENERATIVE ADVERSARIAL NETWORKS (GANS):

2. • GANs - Generative Adversarial Networks
• Introduced by Ian Goodfellow and other researchers from the University of
Montreal in 2014
WHAT ARE GANs?
• It is a neural network that belongs to the set of generative AI
• They are capable of generating new content that has never
been generated before
• GANS fall under the category of Unsupervised Learning
• It can be applied to various domains such as images, music, speech,
and writing
• The architecture consists of two neural networks: the
Generator and the Discriminator

3. THE EVOLUTION OF GANs
• It was a major breakthrough in the field of Deep Learning and Artificial Neural
Networks.
• Since its publication, it has remained one of the most popular and relevant
topics within the field of Artificial Intelligence and Deep Learning.
• It caused a significant advancement as until then other technologies were not
achieving satisfactory results.
• GANs came as a revolution, being the first to produce high-quality results on
most of the datasets they were trained on.

4. CONVOLUTIONAL NEURAL NETWORK (CNN)
• To understand GANs first we must have little understanding of Convolutional
Neural Networks.
• CNNs are trained to classify images with respect to their labels if an image is
fed to a CNN, it analyzes the image pixel by pixel and is passed through nodes
present in CNN's hidden layers and as an output, it tells what the image is
about or what it sees in the image.
• For example: If a CNN is trained to classify dogs and cats and an image is fed
to this CNN, it can tell whether there is a dog or a cat in that image.
Therefore it can also be called as a classification algorithm

5. HOW GANs WORK
Generator
generate new
instances of data
Discriminator
evaluates the authenticity
of the generated data
GENERATOR
DISCRIMINATOR
Images of the
Training set

6. STEPS OF A GAN
• The Generator considers random values to generate an
image – input.
• The generated image is sent to the Discriminator along
with a stream of images taken from the real dataset (to
use as reference).
• The Discriminator returns the probabilities of the
generated image being real. The result is a number
between 0 (fake) and 1 (authentic).

7. STEPS OF A GAN

8. HOW GANs WORK
• The discriminator network is a standard convolutional network (binary classifier
that labels images as real or false).
• The generator is the opposite: while the standard convolutional classifier takes
the image and downsamples it to return a probability, the generator takes a
random noise vector and upsamples it to generate an image that will be
analyzed by the discriminator.
• Both the discriminator and the generator are trying to optimize a different and
opposite fitness function (loss function).
• During training, the generator and the discriminator will change their behavior.
• They will be one against the other because of the losses.
• This is why we say that the two networks “compete” with each other.

9. HOW GANs WORK
GENERATOR
NETWORK
DISCRIMINATOR
NETWORK

10. TRAINING DISCRIMINATOR
• The discriminator network is trained to
distinguish between real and generated
data, playing an adversarial role with
the generator.
• It uses a loss function to measure the
difference between real and fake data,
guiding the training of the entire GAN
system.

11. TRAINING GENERATOR
Training the generator network to create realistic data from noise.

12. TRAINING GENERATOR

13. HOW GANs WORK

14. HOW GANs WORK

15. TYPES OF GANs
Some examples:
• DCGANs (Deep Convolutional)
• WGANs (Wasserstein)
• SRGANs (Super resolution)
• Pix2Pix (Image-to-image)
• CycleGAN (Cycle Generative)
• StackGAN (Stacked GAN)
• ProGAN (Progressive Growing)
• StyleGAN (Style-Based)
• VQGAN (Vector Quantized)
And other lesser known (but
very interesting) extensions:
• SGAN
• InfoGAN
• SAGAN
• AC-GAN
• GauGAN
• GFP-GAN

16. GANs - APPLICATIONS
Paired image-to-image translation
pix2pix

17. GANs - APPLICATIONS
Super Resolution
upsampling a low-resolution image into a higher resolution
r

18. GANs - APPLICATIONS
Text-to-Image Generation (text2image)
Generating Synthetic Images from textual description

19. GANs - APPLICATIONS
Facial rejuvenation and aging
(and other facial operations such as adding a beard, changing facial expression, etc.)

20. GANs - APPLICATIONS
Generate faces of people that don't exist

21. GANs - APPLICATIONS
Fill missing parts of the image (inpainting)

22. GANs - APPLICATIONS
Generating images of new human poses

23. GANs - APPLICATIONS
Restoration of old images and noise removal

24. Conclusion and Future Directions
Progress - Continued advancements in GAN technology.
Research - Exploring new applications and improvements.
Integration - Integrating GAN into various fields and industries.