GAN Flashcards

1
Q

What are generative models

A

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

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

When training GAN the density estimation is explicitly known T F

A

F

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

GAN principle

A

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

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

Discriminator structure in GAN

A

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

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Gan schema

A

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

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

The generator network of a Gan acts as a …

A

DECODER

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

Guidelines for GAN generator architecture

A

Generator is an upsampling network which implements deep deconvolution.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Guideline for discriminator archietecture in Gan

A

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

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

Goal of the generator network in Gan

A

Try to fool the discriminator by producing real-looking data

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

Role of the discriminator

A

Try to distinguish between real and fake data

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

Gan principle of training

A

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

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Training approach of GAN

A

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

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

Possible tasks for GAN

A

Generating images, image translation

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

How to generate images with a query?

A

Couple LLM and GAN

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

How GAN can be used in biomedicine?

A

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

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

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

A

-ln(prediction)

17
Q

Logloss function when I want the output to be zero

A

-ln(1-pred)

18
Q

Which cost function the discriminator compute for a generated image?

A

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

19
Q

Cost function of the generator in Gan

A

-ln(D(G(z)))

20
Q

Formula of minmax game for GAN

A

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

21
Q

Express mathematically the goal of discriminator

A

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

22
Q

Express mathematically the goal of the generator

A

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

23
Q

Discriminator objective function. Which type of optimization

A

Max_td (everything)

24
Q

Objective function of generator and the kind of optimization

A

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

25
Q

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

A

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

26
Q

Gan algorithm for training

A

At each iteration:

27
Q

GAN recommendations

A

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

28
Q

Define image to image translation

A

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

29
Q

Application of image translation in biomedicine

A

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

30
Q

Which kind of NEt is suggested for image translation

A

Cyclic coherence GAN

31
Q

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

A

F

32
Q

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

A

T

33
Q

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

A

F