Deep Learning for Computer Vision Quiz 2: 4 August 2024 (May 2024 term)

Which one of the below statements is incorrect?

Which one of the following statements is false?

Match the following:

Identify the true statement regarding different loss functions:

In the context of Variational Autoencoders, the idea of variational inference solves the problem of

Which of the following is incorrect for the reparametrization trick used in a Variational AutoEncoder (VAE)?

In an image classification problem, which of the following is the quantity of interest in a generative model?

Which one of the following statements is true?

Which one of the following statements is false?

Consider two 10×10 bounding boxes(one on the upper left and one of the lower right) in an images with an overlapping region of 6 × 6. The Intersection over Union (IoU) between the two boxes is (choose the closest value):

A four-dimensional input vector x = [5, 3,−1, 2] is passed to a hidden layer with a single neuron and an activation function a(.) to obtain z. Assume that the corresponding weights (from input to hidden layer) are [0.4, 0.6, 0.8,−0.2] and bias is 0.4. If the activation function a(.) is Sigmoid, Linear, Indicator function, Softplus, ReLU and Leaky-ReLU, what are the corresponding values of z? (Note that Indicator function returns 1 for a positive input, and 0 otherwise. Softplus is defined as: ln(1 + exp(z)).

Consider a Faster R-CNN object detection framework, which is being trained with a mini-batch size of 2. The RPN produces 512 proposals, out of which 12 and 14 of them are foreground proposals,corresponding to each image in the mini-batch. For a single mini-batch update, the bounding box regression loss will be computed_____________________ times.

The feature embeddings obtained from a neural network for an anchor sample, positive sample and negative sample are as given below:

• Anchor Sample: $[4, 5, 3, 7, 1]$
• Positive Sample: $[8, 5, 6, 7, 1]$
• Negative Sample: $[9, 1, 3, 8, 0]$

The triplet loss given that margin $\alpha = 12$ will be __________
(Hint: Triplet loss uses squared euclidean distance.)

A Siamese network is trained for face verification. The face image to be verified is passed through the first branch and the anchor image is passed through the second branch. There are 73286 parameters in each of the branches. Triplet loss is used to learn the overall Siamese network. The total number of trainable parameters in the network (assuming there are no other layers or modules) are________________(write your answer as a number).

Based on the above data, answer the given subquestions.

z1=_________________

Based on the above data, answer the given subquestions.

z2=____________________

Based on the above data, answer the given subquestions.

z3=_______________

KL divergence is not a distance metric because $\text{KL}(P\|Q)$ is (a) __________ to $\text{KL}(Q\|P)$. And KL divergence is always $\geqslant$ (b) __________.
Suppose $P \sim \text{Bernoulli}(0.8)$ and $Q \sim \text{Bernoulli}(0.2)$ then
(c) $\text{KL}(P\|Q) =$ __________ nats and (d) $\text{KL}(Q\|P) =$ __________ nats.

Hint: You can use these values of natural log: $\ln(2) = 0.69$, $\ln(4) = 1.38$, $\ln(0.6) = -0.510$, $\ln(8) = 2.079$, $\ln(0.4) = -0.91$,
Answer upto two decimal places.

Based on the above data, answer the given subquestions.

Enter the correct answer for (a)
**NOTE:**Enter the exact answer without any space in the beginning or at the end.

KL divergence is not a distance metric because $\text{KL}(P\|Q)$ is (a) __________ to $\text{KL}(Q\|P)$. And KL divergence is always $\geqslant$ (b) __________.
Suppose $P \sim \text{Bernoulli}(0.8)$ and $Q \sim \text{Bernoulli}(0.2)$ then
(c) $\text{KL}(P\|Q) =$ __________ nats and (d) $\text{KL}(Q\|P) =$ __________ nats.

Hint: You can use these values of natural log: $\ln(2) = 0.69$, $\ln(4) = 1.38$, $\ln(0.6) = -0.510$, $\ln(8) = 2.079$, $\ln(0.4) = -0.91$,
Answer upto two decimal places.

Based on the above data, answer the given subquestions.

Enter the correct answer for (b)

KL divergence is not a distance metric because $\text{KL}(P\|Q)$ is (a) __________ to $\text{KL}(Q\|P)$. And KL divergence is always $\geqslant$ (b) __________.
Suppose $P \sim \text{Bernoulli}(0.8)$ and $Q \sim \text{Bernoulli}(0.2)$ then
(c) $\text{KL}(P\|Q) =$ __________ nats and (d) $\text{KL}(Q\|P) =$ __________ nats.

Hint: You can use these values of natural log: $\ln(2) = 0.69$, $\ln(4) = 1.38$, $\ln(0.6) = -0.510$, $\ln(8) = 2.079$, $\ln(0.4) = -0.91$,
Answer upto two decimal places.

Based on the above data, answer the given subquestions.

Enter the correct answer for (c)

KL divergence is not a distance metric because $\text{KL}(P\|Q)$ is (a) __________ to $\text{KL}(Q\|P)$. And KL divergence is always $\geqslant$ (b) __________.
Suppose $P \sim \text{Bernoulli}(0.8)$ and $Q \sim \text{Bernoulli}(0.2)$ then
(c) $\text{KL}(P\|Q) =$ __________ nats and (d) $\text{KL}(Q\|P) =$ __________ nats.

Hint: You can use these values of natural log: $\ln(2) = 0.69$, $\ln(4) = 1.38$, $\ln(0.6) = -0.510$, $\ln(8) = 2.079$, $\ln(0.4) = -0.91$,
Answer upto two decimal places.

Based on the above data, answer the given subquestions.

Enter the correct answer for (d)

Consider generative and discriminative models for a classification task. Generative models learn the data-generating mechanism by modeling class-conditional and prior distributions before making predictions. Discriminative models make predictions based on patterns learned in the training set. Which one of the following is an example of a generative model? Select all that are correct.

Which of the following are true:

Which of the following is true regarding Hard Attention and Soft Attention?