Final

Question 1

If we build a neural network with no activation function, we get a nonlinear classifier. (T or F)

Answer 1

False

Question 2

What does a neuron compute?

Answer 2

A neuron computes a linear function (z = Wx +b) followed by an
activation function.

Question 3

During the backwards pass, each node in the graph receives
_______ gradients and multiplies them by __________ gradients to
compute _________ gradients.

Answer 3

Upstream, local, downstream

Question 4

What is the benefit of using a Pooling Layer in a CNN?

Answer 4

Reduces the dimensionality of each feature map

Question 5

The convolution operation shrinks the matrix of pixels (input image)
only if the size of the filter is greater than 1.

Answer 5

True

Question 6

In _______ layer, every neuron in the previous layer is connected to
each and every neuron in the next layer.

Answer 6

Fully connected

Question 7

Generative Adversarial Networks (GANs) jointly train generator and ____________ with a minimax game.

Answer 7

c. Discriminator

Question 8

Match each of the following models with its corresponding:

• Assign labels to data
• Feature Learning (with labels)

Answer 8

Discriminative Model: Learn a probability distribution p(y|x)

Question 9

Match each of the following models with its corresponding:

• Assign labels, while rejecting outliers
• Generate new data conditioned on input labels

Answer 9

Conditional Generative Model: Learn p(x|y)

Question 10

Match each of the following models with its corresponding:

• Detect outliers
• Feature learning (without labels)
• Sample to generate new data

Answer 10

Generative Model: Learn a probability distribution p(x)

Question 11

What is the problem described below?

Do well on training data, but poorly on validation data due to variance.

Answer 11

Overfitting

Question 12

What is the given definition below?

Label each pixel in the image with a category label. It only cares about pixels.

Answer 12

Semantic Segmentation

Question 13

We developed a segmentation algorithm, and we like to check how good our segmentation prediction is. How can we compare our segmentation prediction to the ground-truth?

Answer 13

Intersection over Union

Question 14

When we visualize the first layer of a deep network, what will we see in the output of these layers?

Answer 14

Low level features such as edges and corners

Question 15

Which visualization technique is used to understand the flow of gradients through different layers of a Convolutional Neural Network (CNN)?

Answer 15

Backpropagation

Question 16

Which of the following techniques is used to highlight the most relevant regions in an inpt image for explaining the predictions of a Convolutional Neural Networks (CNNs)?

Answer 16

Grad-CAM

Question 17

List the SIFT Descriptors for feature detection in order:
1) Normalize rotations

Answer 17

1) Normalize the rotation / scale of the patch

2) Compute gradient at each pixel
3) Divide into sub-patches (here 2x2, actually 4x4)

4) In each sub-patch, compute histogram of 8 gradient directions

5) Describe the patch with 448 = 128 numbers

Question 18

What Components of a Convolutional Network is this?

Answer 18

Full Connected Layer

Question 19

What Components of a Convolutional Network is this?

Answer 19

Activation Function

Question 20

What Components of a Convolutional Network is this?

Answer 20

Convolutional Layers

Question 21

What Components of a Convolutional Network is this?

Answer 21

Pooling Layer

Question 22

What Components of a Convolutional Network is this?

Answer 22

Normalization

Question 23

List the steps for Training a Convolutional Network and what does each step do?

Answer 23

1) Download big datasets: acquire and download large dataset

2) Design CNN architecture: Make a CNN from scratch or modify existing CNN’s like Alexnet, Vgg, or googlenet

3) Initialize Weights: set weights to desired initial values

4) For t = 1 to T:
1. Form minibatch: form a small
batches of data from the training
set used to trained the model.

        2. compute loss + gradient: The 
        model's prediction
        3. Update Weights

Question 24

What Is Morphology?

Answer 24

Morphological image processing (or
morphology) describes a range of image
processing techniques that deal with the shape
(or morphology) of features in an image

Question 25

Morphological operations

Answer 25

typically applied
to remove imperfections introduced during
segmentation, and so typically operate on bi-
level images
17

Question 26

two basic morphological

Answer 26

erosion and dilation

Question 27

More interesting morphological operations can
be performing combinations of
erosions and dilations
The most widely used of these compound
operations are:

Answer 27

– Opening
– Closing

Question 28

Answer 28

• Roberts filter
• Prewitt filter
• Sobel filter

Question 29

Name a voting scheme

Answer 29

Hough transform

Question 30

In backward propagation what happens in the backward pass?

Answer 30

During the backward pass, each node in the graph receives upstream gradients and multiplies them by local gradients to compute downstream gradients

Question 31

used for Circle Detection

Answer 31

Hough transforms

Question 32

region:
no change in
all directions

Answer 32

“flat

Question 33

no change along
the edge
direction

Answer 33

edge

Question 34

significant
change in all
directions

Answer 34

corner

Question 35

SIFT Descriptors

Answer 35

1. Normalize the rotation / scale of the patch
2. Compute gradient at each pixel
3. Divide into sub-patches (here 2x2, actually 4x4)
4. In each sub-patch, compute histogram of 8
   gradient directions
5. Describe the patch with 448 = 128 numbers

Question 36

What happens in backpropagation during backward pass?

Answer 36

During the backward pass, each node in the graph receives upstream gradients and multiplies them by local gradients to compute downstream gradients

Question 37

Fill in the blanks

Answer 37

Question 38

What are the different machine vision tasks?

Answer 38

• Classification - no spatial extent
• semantic segmentation - no objects, just pixels
• object detection:

Question 39

What is a Discriminative Model

Answer 39

Assign labels to data, Feature Learning with labels

Question 40

What is a Generative Model

Answer 40

Detect outliers, Feature Learning without labels

Question 41

What is a Conditional Generative Model

Answer 41

Assign labels, while rejecting outliers

Question 42

What is GANs?

Answer 42

Generative Adversarial Networks

Question 43

What does GANs do?

Answer 43

Jointly train generator and discriminator with minimax game