Quiz 7: CS 7643 Deep Learning | Questions with Verified Answers | 100% Correct| Latest 2025/2026 Update - Georgia Institute of Technology.

Quiz 7: CS 7643 Deep Learning | Questions
i,- i,- i,- i,- i,- i,- i,- i,-




with Verified Answers | 100% Correct| Latest
i,- i,- i,- i,- i,- i,- i,-




2025/2026 Update - Georgia Institute of i,- i,- i,- i,- i,- i,-




Technology.

Tractable Density Estimation
i,- i,- i,-i,- i,- - Explicit Density
i,- i,-




- Simplify joint distribution into factorized model of simpler
i,- i,- i,- i,- i,- i,- i,- i,- i,-



components and optimize with respect to simpler components
i,- i,- i,- i,- i,- i,- i,-




- PixelRNN/PixelCNN
i,-




Approximate Density Estimation i,- i,- i,-i,- i,- - Explicit Density
i,- i,-




- Learn distributions that approximate true joint distribution
i,- i,- i,- i,- i,- i,- i,-




- Variational Autoencoder
i,- i,-




Implicit Density Estimation
i,- - We don't model density itself, i,- i,-i,- i,- i,- i,- i,- i,- i,- i,-



but can sample from the distribution
i,- i,- i,- i,- i,-




- GAN
i,-




Factorized Models for Images Downsides
i,- i,- i,- i,- i,-i,- i,- 1. Slow sequential
i,- i,- i,-



generation process i,-




2. Only considers few context pixels
i,- i,- i,- i,- i,-

, Discriminative vs. Generative Models i,- i,- i,- i,-i,- i,- 1. Discriminative
i,- i,-



models model P(y|x) i,- i,-




- NN, SVM, etc.
i,- i,- i,-




- No way to handle unreasonable outputs (i.e. if trained on cats
i,- i,- i,- i,- i,- i,- i,- i,- i,- i,- i,- i,-



and dog images, will always output a label of cat or dog)
i,- i,- i,- i,- i,- i,- i,- i,- i,- i,- i,-




2. Generative Models model P(x)
i,- i,- i,- i,-




- Can parameterize our model as P(x, theta) and use Maximum
i,- i,- i,- i,- i,- i,- i,- i,- i,- i,- i,-



Likelihood Estimation to optimize the parameters i,- i,- i,- i,- i,-




- Called generative because they can generate samples
i,- i,- i,- i,- i,- i,- i,-




- Can "reject" unreasonable inputs as being too unlikely
i,- i,- i,- i,- i,- i,- i,- i,-




- Feature Learning without labels
i,- i,- i,- i,-




Explicit vs Implicit Density Explicit: Explicitly define and solve
i,- i,- i,- i,-i,- i,- i,- i,- i,- i,- i,-



for Pmodel(x|θ), given an explicit likelihood that can be maximized
i,- i,- i,- i,- i,- i,- i,- i,- i,-




Implicit: Learn a parameterized generation model that can sample
i,- i,- i,- i,- i,- i,- i,- i,- i,-



from the joint distribution Pmodel(x|θ) without explicitly defining
i,- i,- i,- i,- i,- i,- i,- i,-



Pmodel(x|θ)