Scribd
Upload
31 views21 pages

Generative Adversial Network

A detailed presentation on Generative Advanced Neural networks and their applications!

Uploaded by

Muhammad Ali
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
31 views21 pages

Generative Adversial Network

A detailed presentation on Generative Advanced Neural networks and their applications!

Uploaded by

Muhammad Ali
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 21

Generative

Adversarial
Network (GAN)
Get Started
Group Members

 Haider Ali (2766)

 Hamza Shakeel (2760)

 Jawad ul Hassan (2752)

 Muhammad Ali (2757)


Agenda

 Introduction

 Working of GANs

 Types of GANs

 Application of GANs

 Conclusion
n

Introduction to
Generative
Adversarial Networks
Generative Adversarial Networks (GANs) are a powerful deep
learning framework with transformative applications.

By Jawad Ul Hassan
n

Introduction
Definition:

Generative Adversarial Networks (GANs) are a type of Artificial Intelligence (AI) model used to create
new, realistic data by learning patterns from existing data

Concept :
 Two Models in a Game:

• Generator: Creates fake data (like a forger making fake art).


• Discriminator: Identifies real vs. fake data (like an art critic).

 Learning Process:

• Both models compete and improve over time:


• The Generator gets better at creating realistic data.
• The Discriminator gets better at spotting fakes.
• Output: GANs can generate realistic images, videos, or even sounds
n

Visual Representation
n

Working of GANs

Generative Adversarial Networks, or GANs, are comprised of two


neural networks: a generator and a discriminator.

By Haider Ali
Working of GANs : Generative Adversarial Networks (GANs) consist of
two neural networks: a generator that creates fake data
and a discriminator that distinguishes between real and
fake data. The generator improves over time, producing
more realistic outputs.

01 Generator 02 Discriminator
The generator in a GAN creates The discriminator in a GAN
fake data resembling the real distinguishes between real and
dataset and improves by trying fake data, improving its accuracy
to fool the discriminator into by identifying whether the input
believing it's real. is real or generated.

03 Discriminator 04 Generator Loss


Loss Loss measures the
Discriminator Generator Loss measures how well
discriminator's ability to the generator fools the
distinguish real from fake data, discriminator, penalizing it when
penalizing incorrect
classifications to improve its outputs are correctly identified
accuracy. as fake, encouraging more realistic
data.
Figure

Discrimanto
Real r Loss
Samples
Images

Discriminator

Generator
Loss
Random Generator
Input
Example

Deep fake:
•GANs are used to create hyper-realistic images or videos where a
person's face is swapped with another. The generator creates fake
faces, while the discriminator distinguishes between real and
generated faces. The process continuously improves deepfakes,
making them harder to identify as fake. This technology has
applications in entertainment but also raises concerns related to
misinformation and privacy
Types of Generative
Adversarial Networks
Various GAN architectures have emerged, each offering unique
strengths for specific tasks.

By Hamza Shakeel
Types of GANs :
1) Vanilla GAN

- The original GAN model proposed by Ian


Goodfellow in 2014.
- Consists of a simple Generator and Discriminator.
- Uses a minimax game to train both networks.
- Applications: Image generation, data
augmentation.
2) Deep Convolutional GAN (DCGAN)

- Introduces convolutional layers to GANs.


- Provides better stability and quality in image
generation.
- Uses strided convolutions and batch
normalization.
- Applications: High-quality image generation.
3) Conditional GAN (CGAN)

- Allows control over the output by


conditioning on labels.
- The generator and discriminator receive an
additional input (e.g., class labels).
- Applications: Image-to-image translation,
text-to-image generation.
4) Super-Resolution GAN (SRGAN)

- Focuses on generating high-resolution


images from low-resolution inputs.
- Uses perceptual loss to enhance image
details and realism.
- Applications: Enhancing low-resolution
photos, medical imaging, satellite imagery.
Applications of GANs

Unlocking the potential of GANs across various fields, from


realistic image generation to medical innovation.

Muhammad Ali Sardar


Applications of GANs

Image and Video Healthcare


DeepFakes, style transfer, Synthetic medical images, drug
super-resolution discovery

Gaming & Simulation Audio & Text


Realistic environments, Speech synthesis, music creation
autonomous vehicle training
Image and Video Applications of GANs

Super-Resolution DeepFakes Style Transfer


Enhancing low-res images to high Realistic face swapping, ethical Transforming photos into artwork
quality considerations
GANs in Healthcare
Synthetic Medical Images
Drug Discovery

GANs generate synthetic GANs design new drug molecules,


images of rare diseases, speeding development and
improving AI model training and reducing costs.
diagnosis accuracy.
Synthetic data addresses data This accelerates drug candidate
scarcity for rare conditions. design and clinical trials.
GANs in Gaming & Simulation
Gaming Applications Simulation Applications
GANs train autonomous vehicles in diverse scenarios (road
GANs create realistic game environments and characters,
enhancing immersion. They create virtual environments for training AI,
Players experience more believable and engaging gameplay
simulating crowds and complex physics.
Conclusion
Transformative Ethical
technology considerations
Wide-reaching Misuse in DeepFakes,
applications across mode collapse
industries.
High-quality data Future prospects
generation
Innovations in AI-driven
Boosting creativity
and innovation content creation,
simulation, and
personalization.

You might also like