Thanks to visit codestin.com
Credit goes to www.scribd.com

Scribd
Upload
36 views11 pages

Generative Adversarial Networks

Generative Adversarial Networks (GANs) are machine learning models introduced by Ian Goodfellow in 2014 that generate hyper-realistic data through a competition between a Generator and a Discriminator. They have applications in image generation, data augmentation, and super-resolution, but face challenges like mode collapse and high computational costs. Ongoing research aims to improve the quality and diversity of outputs, expanding their use beyond images to audio and video.

Uploaded by

lynie.caderao
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
36 views11 pages

Generative Adversarial Networks

Generative Adversarial Networks (GANs) are machine learning models introduced by Ian Goodfellow in 2014 that generate hyper-realistic data through a competition between a Generator and a Discriminator. They have applications in image generation, data augmentation, and super-resolution, but face challenges like mode collapse and high computational costs. Ongoing research aims to improve the quality and diversity of outputs, expanding their use beyond images to audio and video.

Uploaded by

lynie.caderao
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/ 11

YMIR LOUIE JASPE

JIM ISAAC MORGIA


BSCPE 4B

GENERATIVE ADVERSARIAL
NETWORKS
GANS
Introduction to GANs How GANs Work Architecture of GANs Implementation Steps

Applications of GANs Challenges and Limitations


Advance Techniques Conclusion

CONTENTS
Introduction to GANs
•Definition: GANs are a class of machine
learning models designed to generate hyper-
realistic data such as images.
•Inventor: Introduced by Ian Goodfellow in
2014.
•Key Idea: GANs consist of two neural networks
(Generator
CONTENTS & Discriminator) competing in a
zero-sum game.
How GANs Work
• Generator: Creates fake data (images) from
random noise.
• Discriminator: Determines if the data is real
or fake.
• Training Process: Both networks are trained
simultaneously until the Generator produces
data indistinguishable from real data.
CONTENTS
Architecture of GANs
•Generator: Uses input noise to
Generator
create fake data (e.g., images).
•Discriminator: Classifies data Fake
Image
as real or fake. Loss Functions
•Loss Functions: Binary Cross-
Entropy or Wasserstein Loss for
Discriminator
stability and convergence.
Real
Image

Domai
n
Implementation Steps
1.Collect Data: Training on a large dataset of real
images.
2.Design Network: Build the architecture of the
Generator and Discriminator using frameworks like
TensorFlow or PyTorch.
3.Train Model: Adversarial training using loss
functions to balance the performance of both
networks.
4.Fine-tuning: Adjust hyperparameters like
Applications of GANs
•Image Generation: Generating art, face
generation (e.g., "This Person Does Not Exist").
•Data Augmentation: Creating synthetic training
data for various machine learning tasks.
•Image-to-Image Translation: Converting
sketches to real images or style transfer.
•Super-Resolution: Enhancing image quality by
generating higher-resolution versions.
Advance Techniques

•Progressive Growing: Improves


training stability by gradually increasing
image resolution.
•Conditional GANs (cGANs): Generate
images based on specific conditions (e.g.,
generating images of a dog breed).
Challenges and Limitations

•Mode Collapse: GAN produces limited


diversity in outputs.
•Training Instability: GANs can be difficult
to train and balance.
•Computational Costs: Require significant
resources to train effectively.
Conclusion

•Impact: GANs are a breakthrough in


unsupervised learning and have
revolutionized how machines generate
realistic images.
•Future Scope: Ongoing research is
enhancing the quality and diversity of
generated outputs with applications
extending beyond imaging to audio, video,
THANK YOU
YMIR LOUIE JASPE
JIM ISAAC MORGIA
BSCPE 4B

You might also like