Generative Adversarial Network

Applied Deep Learning

Guest Lecture by Hung-yi Lee

Generative Adversarial Network

May 19th, 2020

Three Categories of GAN

1. Typical GAN

?0.3

0.1

?

0.9

random vector

Generator

image

2. Conditional GAN

blue eyes,

red hair,

short hair

paired data

��Girl with

red hair��

Generator

text

image

3. Unsupervised Conditional GAN

domain x

domain y

x

y

Generator

Photo

unpaired data

Vincent van

Gogh��s style

Generative Adversarial Network

(GAN)

? Anime face generation as example

vector

image

Generator

image

Discriminator

score

high

dimensional

vector

Larger score means real,

smaller score means fake.

Algorithm

? Initialize generator and discriminator

? In each training iteration:

G

D

Step 1: Fix generator G, and update discriminator D

sample

1

Database

generated

objects

1

1

D

0

vector

vector

vector

vector

randomly

sampled

1

Update

0

0

G

0

Fix

Discriminator learns to assign high scores to real objects

and low scores to generated objects.

Algorithm

? Initialize generator and discriminator

? In each training iteration:

G

D

Step 2: Fix discriminator D, and update generator G

Generator learns to ��fool�� the discriminator

hidden layer

vector

NN

Generator

update

large network

Backpropagation

Discriminator

fix

0.13

 ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download