GAN

Question 1

What are generative models

Answer 1

Unsupervised method to build a model explaining data. Given training data they generate new instances extracted from modeled distribution

Question 2

When training GAN the density estimation is explicitly known T F

Answer 2

F

Question 3

GAN principle

Answer 3

Model distribution is learnt from training distribution to generate data through competitive two player game

Question 4

Discriminator structure in GAN

Answer 4

Encoder (AE or CNN) and classifier (logistic neuron)

Question 5

Gan schema

Answer 5

We get real sample from dataset and a generated sample by the generator which is fed with a random variable. The two samples are alternatively introduced in the net of the discriminator which find who is fake

Question 6

The generator network of a Gan acts as a …

Answer 6

DECODER

Question 7

Guidelines for GAN generator architecture

Answer 7

Generator is an upsampling network which implements deep deconvolution.

Question 8

Guideline for discriminator archietecture in Gan

Answer 8

Use convolutional network performing classification. Some authors proposed also Fullyconnected FF

Question 9

Goal of the generator network in Gan

Answer 9

Try to fool the discriminator by producing real-looking data

Question 10

Role of the discriminator

Answer 10

Try to distinguish between real and fake data

Question 11

Gan principle of training

Answer 11

The generator must be able to successfully trick the discriminator so that we are generating data that look like data from the training set

Question 12

Training approach of GAN

Answer 12

Estimating jointly teta-d e teta-g in the so-called minimax game

Question 13

Possible tasks for GAN

Answer 13

Generating images, image translation

Question 14

How to generate images with a query?

Answer 14

Couple LLM and GAN

Question 15

How GAN can be used in biomedicine?

Answer 15

Synthetic data generation for augmentation in training setup
Image tot image translation
Protein and gene synthetic data bank

Question 16

Log-loss function when I want the output to be one

Answer 16

-ln(prediction)

Question 17

Logloss function when I want the output to be zero

Answer 17

-ln(1-pred)

Question 18

Which cost function the discriminator compute for a generated image?

Answer 18

-ln(1-D(G(z)))

Question 19

Cost function of the generator in Gan

Answer 19

-ln(D(G(z)))

Question 20

Formula of minmax game for GAN

Answer 20

Expectation values are used.
Min_tg max_td (E_x_pdata *ln(D(x)) +E_z_pmodel * ln(1-D(G(z)))

Question 21

Express mathematically the goal of discriminator

Answer 21

Maximize objective such that D(x) is close to one and D(G(z)) is close to 0

Question 22

Express mathematically the goal of the generator

Answer 22

Wants to minimize objective such that D(G(z)) is close to 1

Question 23

Discriminator objective function. Which type of optimization

Answer 23

Max_td (everything)

Question 24

Objective function of generator and the kind of optimization

Answer 24

Gradient descent
Min_tg(E_z(ln(1-Dtd(G5g(z))))

Question 25

What is the problem of gradient descent on generator and how to solve

Answer 25

When sample is fake gradient is relatively flat while gradient signal is high mainly in the region where sample is already good.. We maximize instead the likelihood of discriminator being wrong

Question 26

Gan algorithm for training

Answer 26

At each iteration:

Question 27

GAN recommendations

Answer 27

Replace any pooling layers with strided convolution(discriminator) and fractional strided convolutions(gen)
Noted of FC LAYERS
Use ReLU in All layers of generator except output (use tank)
Use LeakyReLU in the discriminator for all layers

Question 28

Define image to image translation

Answer 28

Transform images belonging to domain A into different domain B by transferring the typical characteristics of domain B

Question 29

Application of image translation in biomedicine

Answer 29

Radiotherapy: re-planning and adaptation during treatment session
Diagnostic: use in medical centers with more limited resources

Question 30

Which kind of NEt is suggested for image translation

Answer 30

Cyclic coherence GAN

Question 31

GAN needs a priori density function of the generator t or f

Answer 31

F

Question 32

Gan learns how to generate from training data through 2-players game t or f

Answer 32

T

Question 33

GAN is more stable to train respect to traditional networks T or F

Answer 33

F