Computer vision Intern position - Set2

1. (Multiple Choice) Which of the following is not typically an output of a semantic segmentation
process?
A) A label for each pixel
B) A mask for each detected object
C) A classification score for the whole image
D) Boundaries of detected objects
Ans. C) A classification score for the whole image

2. (True/False) A higher learning rate always speeds up the training of a deep learning model
without any adverse effects.
Ans. Flase
3. (Fill-in-the-Blank) The term__________refers to the modification of a neural network's
architecture to reduce the complexity and computational cost.
Ans. Pruning
4. (Short Answer) What is the primary benefit of using batch normalization in a deep learning
model?
Ans. The primary benefit of using batch normalization in a deep learning model is that it helps
stabilize and accelerate the training process by normalizing the input to each layer, which
reduces issues related to vanishing or exploding gradients. This allows for the use of higher
learning rates and can lead to faster convergence and better overall performance.

5. (Multiple Choice) In the context of an image, 'depth' typically refers to:

A) The number of color channels in the image
B) The distance of objects from the camera
C) The focus level of the camera
D) The resolution of the image
Ans. A) The number of color channels in the image

6. (Fill-in-the-Blank)___________learning is a type of machine learning where the algorithm

discovers the patterns directly from the data without any labeled examples.
Ans.
Unsupervised
7. (True/False) The main goal of principal component analysis (PCA) is to classify the data into
different categories.
Ans. False

8. (Multiple Choice) Which of the following is a common use case for a recurrent neural network
(RNN)?
A) Image classification
B) Time series prediction
C) Static object detection
D) Edge detection
Ans. B) Time series prediction

20JG1A4219 GEDDAM RUPA

 9. (Short Answer) Explain the term 'backbone network' in the context of computer vision.
Ans.
n computer vision, a 'backbone network' refers to the initial part of a deep neural network architecture
that is responsible for extracting low-level features from an input image. It forms the foundation upon
which additional layers and modules are built for more complex tasks such as object detection or
segmentation. The backbone network typically consists of convolutional layers that gradually learn
hierarchical representations of the image, starting from edges and textures to more abstract features
like shapes and structures. These learned features are then passed on to subsequent layers for
further processing and interpretation. The choice of backbone network can significantly influence the
performance of a computer vision model. Popular backbone networks include architectures like VGG,
ResNet, and MobileNet.

10. (Fill-in-the-Blank)_________is the measure of the actual output compared to the predicted
output in a regression model.
Ans. Residual

11. (True/False) In a convolutional neural network, the filter sizes in the convolutional layers
typically decrease as you go deeper into the network.
Ans. False
12. (Multiple Choice) What is the primary function of the softmax activation function?
A) To normalize the output to be between 0 and 1
B) To introduce non-linearity into the model
C) To reduce the dimensionality of the input
D) To amplify the strongest signal in the input
Ans. A) To normalize the output to be between 0 and 1

13. (Short Answer) What is the difference between a generative and a discriminative model?
Ans.
A generative model learns the joint probability distribution of the input features and their
corresponding labels. It can generate new samples that are similar to the training data.
Examples of generative models include Variational Autoencoders (VAEs) and Generative
Adversarial Networks (GANs).
A discriminative model, on the other hand, learns the conditional probability distribution of the
labels given the input features. Its primary task is to classify or differentiate between different
classes. Logistic regression and support vector machines are examples of discriminative
models.
In summary, generative models learn how data is generated, while discriminative models
focus on the boundary between different classes.

14. (Fill-in-the-Blank) A________is a model that is composed of multiple layers of artificial

neurons that can learn representations of data.
Ans.
Neural network
15. (True/False) In image processing, 'filtering' refers exclusively to the removal of noise from
an image.
Ans. False
16. (Multiple Choice) What is the main purpose of using stride in a convolutional neural network?
A) To increase the number of filters
B) To control the overlap between receptive fields
C) To reduce the computational load
D) To increase the sensitivity to noise
20JG1A4219 GEDDAM RUPA
Ans. B) To control the overlap between receptive fields

17. (Fill-in-the-Blank) The_________algorithm is an optimization method that can be used to

speed up learning and perhaps even lead to better final performance.
Ans.
Adam
18. (Short Answer) Describe the concept of 'data leakage' in machine learning.
Ans.
Data leakage in machine learning refers to a situation where information from the
validation or test set is inadvertently used during the training process. This can lead to overly
optimistic performance estimates and models that fail to generalize well to new, unseen data.
There are two main types of data leakage:
Leakage through features: This occurs when information from the validation or test set is
included as a feature during training. For example, using the target variable itself or related
statistics from the validation or test set as features can introduce data leakage.
Leakage through time: In time series data, it's important to respect the temporal order of
observations. Using future information to predict the past, forexample, can lead to models
that perform unrealistically well in practice but fail when applied to new data.
Preventing data leakage is crucial for building robust and reliable machine learning models. It
involves careful preprocessing, feature engineering, and ensuring that information from the
validation and test sets is not inadvertently used during training.

19. (True/False) An image with higher contrast is always better for computer vision tasks.
Ans. False
20. (Multiple Choice) Which of the following is a disadvantage of using deep learning models?
A) They require large amounts of data
B) They are too fast
C) They are too simple
D) They do not require feature engineering
Ans. A) They require large amounts of data

21. (Fill-in-the-Blank)_________is the process of identifying and correcting inconsistencies in

data to improve its quality.
Ans.
Data cleaning
22. (Short Answer) What is the role of a loss function in machine learning?
The role of a loss function in machine learning is to quantify how well the predictions of a
model match the true labels or target values in the training data. It provides a measure of the
model's performance, with the goal of minimizing this loss during the training process. The
loss function guides the optimization algorithm in adjusting the model's parameters (such as
weights in a neural network) to improve its predictive accuracy. Different tasks and models
may require different types of loss functions, tailored to the specific problem at hand, such as
mean squared error for regression tasks or categorical cross-entropy for classification tasks.

23. (True/False) Convolutional neural networks are primarily used for vector data.
Ans. False

24. (Multiple Choice) What is the term for the process of adjusting the parameters of a model
20JG1A4219 GEDDAM RUPA
 in response to the error of the output?
A) Forward propagation
B) Backpropagation
C) Feedforward
D) Feedback loop
Ans. B) Backpropagation

25. (Fill-in-the-Blank) In the context of computer vision, the term________refers to the process of
identifying specific human behaviors in video data.
Ans.
Action recognition

20JG1A4219 GEDDAM RUPA

20JG1A4219 GEDDAM RUPA
20JG1A4219 GEDDAM RUPA