Introduction to
Artificial Intelligent
(AI)
Generative
Adversarial Network
Generative
Adversarial
Networks
• GAN is a deep learning architecture.
• GAN trains two neural networks to compete against
each other to generate more authentic new data from
a given training dataset.
2
Generative Adversarial
Networks
3
The discriminator
• The Discriminator is a classification
model.
• The discriminator is also used to classify
whether the output from the generator
is real or fake.
• If the discriminator performs
significantly better than the generator
Vanishing Gradients
• Discriminator loss function
Real sample Fake sample
4
The Generator
• The Generator is a generative
model.
• This model generates samples that
are intended to resemble the
samples from our training set.
• the generator is always trying to
find the one output that seems
most plausible to the
discriminator.
• Generator loss function
5
Training GAN
1. Train the Discriminator
2. Set the Discriminator as
non-trainable
3. Train the Generator
4. Repeat again and again
6
Vanilla GAN
7
Vanilla GAN
8
Vanilla GAN
9
Deep convolution GAN
• Deep Convolutional GAN was published in 2016 by
Radford et al.
• DCGAN emphasized using deeper architectures with
Up-sampling layer for the Generator to create high
resolution image
DCGAN generator
10
architecture
Deep convolution GAN
DCGAN generator
architecture
11
Deep convolution GAN
12
Conditional GAN
• Conditional GANs (CGANs)
are the class of GANs that
provide the control needed
to generate a specific
output.
• CGANs work by training
the generator model to
generate fake samples
conditioned on specific
characteristics of the
output required
13
Conditional GAN
14
Conditional GAN
15
Wasserstein GAN
• W-GANs treat the discriminator as a critic
• W-GAN discriminator (or critic) generates a score to
inform the generator about the realness or fakeness of
the input image
16
Wasserstein GAN
17
Time-Series GAN
18
Time-Series GAN
Popular Dataset for Time-series
GAN
19
Time-Series GAN
RCGAN architecture with conditional input c, input data x and latent
variable z
(2017)
20
Time-Series GAN
RCGAN architecture with conditional input c, input data x and latent
variable z
(2017)
21
22
