CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
Correct 30
Incorrect 00
CS7643 Quiz 5 - P2 exam, tested questions
with correct and verified answers,100%
pass.
Term
Give this one a go later!
means and covariance
[Type here]
CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
,CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
E[log p(x|z)] - D_KL(q(z|x) || p(z))
The expectation with respect to latent features z (output of decoder) minus
the KL divergence between the gaussians encoder and z prior.
By maximizing we have a guaranteed lower bound for VAE
- Not very good
- High gradient when D(G(z)) is high (discriminator is wrong)
- We want it to improve when samples are bad (discriminator is right)
Generator is minimizing
Discriminator is maximizing
min_G max_D V(D,G) = E[log D(x)] + E[log(1 - D(G(z))]
Note this does change after fixing the gradient issue
The discriminator is maximizing this objective by pushing D(x) to 1 and D(G(z)) to
0, the generator is minimizing it by pushing D(G(Z)) to 1.
KL Divergence is a distance measure between probability distributions, and
always >=0
If two distributions are the same KL Divergence = 0, if different, unbounded
Don't know?
2 of 30
Definition
Take Z1 that is one dimension of your latent space, Z2 which is
another dimension, and then interpolate between them. E.g. Z1 (0,
[Type here]
CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
,CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
0.1, 0.2, ... 0.9, 1) and same for Z2 and then sample for all
You should see some smoothness in the dimensionality of Z. For the
[Type here]
CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
, CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
example below, Z1 seems to correspond to smiles, Z2 more related
to the orientation of the face
Give this one a go later!
Walking over VAE Latent
Approximate Density Estimation
Space
Mode Collapse Difficulty in GANs
Don't know?
3 of 30
Definition
1. A generator transforms a vector of random numbers to an image
p(x)
2. A discriminator takes p(x) and a series of fake images and attempts
to determine if it's real or fake
Generator: Update weights to improve realism
Discriminator: Update weights to better discriminate
Compete against each other in a minimax game
Trained with minibatch gradient descent.
[Type here]
CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
Correct 30
Incorrect 00
CS7643 Quiz 5 - P2 exam, tested questions
with correct and verified answers,100%
pass.
Term
Give this one a go later!
means and covariance
[Type here]
CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
,CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
E[log p(x|z)] - D_KL(q(z|x) || p(z))
The expectation with respect to latent features z (output of decoder) minus
the KL divergence between the gaussians encoder and z prior.
By maximizing we have a guaranteed lower bound for VAE
- Not very good
- High gradient when D(G(z)) is high (discriminator is wrong)
- We want it to improve when samples are bad (discriminator is right)
Generator is minimizing
Discriminator is maximizing
min_G max_D V(D,G) = E[log D(x)] + E[log(1 - D(G(z))]
Note this does change after fixing the gradient issue
The discriminator is maximizing this objective by pushing D(x) to 1 and D(G(z)) to
0, the generator is minimizing it by pushing D(G(Z)) to 1.
KL Divergence is a distance measure between probability distributions, and
always >=0
If two distributions are the same KL Divergence = 0, if different, unbounded
Don't know?
2 of 30
Definition
Take Z1 that is one dimension of your latent space, Z2 which is
another dimension, and then interpolate between them. E.g. Z1 (0,
[Type here]
CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
,CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
0.1, 0.2, ... 0.9, 1) and same for Z2 and then sample for all
You should see some smoothness in the dimensionality of Z. For the
[Type here]
CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
, CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
example below, Z1 seems to correspond to smiles, Z2 more related
to the orientation of the face
Give this one a go later!
Walking over VAE Latent
Approximate Density Estimation
Space
Mode Collapse Difficulty in GANs
Don't know?
3 of 30
Definition
1. A generator transforms a vector of random numbers to an image
p(x)
2. A discriminator takes p(x) and a series of fake images and attempts
to determine if it's real or fake
Generator: Update weights to improve realism
Discriminator: Update weights to better discriminate
Compete against each other in a minimax game
Trained with minibatch gradient descent.
[Type here]
CS7643 Quiz 5 - P2 exam, tested questions with correct and verified answers,100% pass.
