Scribd
Upload
14 views4 pages

245008-23CS2902 - Deep Learning

Uploaded by

josphineleela ramakrishnan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
14 views4 pages

245008-23CS2902 - Deep Learning

Uploaded by

josphineleela ramakrishnan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 4

Question Paper Code: 245008

Register No:

PANIMALAR ENGINEERING COLLEGE


An Autonomous Institution, Affiliated to Anna University
CHENNAI - 600 123
M.E. DEGREE END SEMESTER EXAMINATIONS NOV / DEC 2024
Third Semester
M.E. – Computer Science and Engineering
23CS2902 – DEEP LEARNING
(Regulation 2023)
Maximum Marks: 100 Duration: 3 Hours
Course Outcome:
CO1: Understand the role of Deep learning in Machine Learning Applications.
CO2: To get familiar with the use of Tensor Flow/ Keras in Deep Learning Applications.
CO3: To design and implement Deep Learning Applications.
CO4: Critically Analyse Different Deep Learning Models in Image Related Projects.
CO5: To design and implement Convolutional Neural Networks.
CO6: To know about applications of Deep Learning in NLP and Image Processing.
Bloom’s Level: 1 - Remembering, 2 - Understanding, 3 - Applying, 4 - Analyzing, 5 - Evaluating 6 - Creating.
Answer All the Questions
Bloom’s Course Marks
PART A (10 x 2 = 20 Marks) Level Outcome Allotted

1. Distinguish between scalar and vector. [BL2 [CO1] [2]


]
2. Why neural networks are considered as universal function [BL3 [CO1] [2]
approximators? ]
3. How does early stopping prevent overfitting in neural networks? [BL3 [CO1] [2]
]
4. State the role of regularization in machine learning models. [BL1 [CO1] [2]
]
5. What is feature extraction? [BL1 [CO5] [2]
]
6. Why CNN is most preferred for image data? [BL3 [CO5] [2]
]
7. For the scenario, “To predict a missing word in a sequence you [BL4 [CO4] [2]
want to look at both the left and right”. Suggest a suitable RNN ]
model to solve this task.
Justify why?
8. What is Bidirectional LSTM? [BL1 [CO4] [2]
]
9. List out the challenges faced while working in object tracking [BL1 [CO3] [2]
algorithms. ]
10. How does a recurrent neural network (RNN) differ from a [BL3 [CO6] [2]
feedforward neural network in NLP applications? ]

PART B (3 x 15 = 65 Marks)

11.a i. Illustrate the perceptron training procedure and the delta [BL2 [CO1] [8]
rule. ]
ii. Define non-linearity issue in a basic XOR-problem. Using [5]
Question Paper Code: 245008

suitable Perceptron classification model, solve this non-


linearity issue.
OR
Question Paper Code: 245008

11.b i. Identify the algorithm that allows us to learn a function or [BL3] [CO1] [8]
relationship from a given set of continuous data. Elaborate it
with its advantages and disadvantages
ii. Construct a simple linear regression model using Python. [5]

i. How does batch normalization help in improving convergence speed, [BL3] [CO1] [6]
stabilizing the training process, and addressing the issue of
internal covariate shift?
ii. Discuss the impact of activation functions on the [7]
expressiveness of neural networks. Provide a detailed
analysis of how different activation functions influence the
capacity of a network to approximate complex functions.
OR
i. Characterize regularization. Explain how L1 and L2 [BL2] [CO1] [6]
regularization techniques prevent overfitting in the model.
ii. Provide a detailed description of how RMSprop optimizer [7]
addresses the challenges of vanishing and exploding
gradients.

i. Assume that your task is to classify the images. But the size of the [BL3] [CO5] [8]
dataset in your hand is small. Which kind of algorithm is suitable
to solve this problem and elaborate it?
ii. Sketch the implementation of the above algorithm using Python. [5]
OR
Consider the following 5X5 image and 3X3 kernel. [BL3] [CO5] [13]
3 3 2 1 0 0 1 2
0 0 1 3 1 2 2 0
3 1 2 2 3 0 1 2
2 0 0 2 2
2 0 0 0 1
Apply CNN algorithm with stride=1 and max pooling operation and find out
the resultant convolved map.

Demonstrate the error back propagation stages of the RNN network with [BL2] [CO4] [13]
an input layer (x), a hidden layer (h), and an output layer (y) based on the
loss function (L) mean square error.
OR
Explain the sequence to sequence prediction problem using an [BL2] [CO4] [13]
encoder decoder model with a neat architecture.

Identify and elaborate the building blocks for implementing the scene [BL3] [CO3] [13]
Understanding task with the integration of CNN and RNN models for
generating deep captions for the objects present in the scene.
OR
Apply an object detector model for labeling the objects with accurate [BL3] [CO3] [13]
bounding box. Narrate the stages involved in detection of objects.
Question Paper Code: 245008

PART C (1 x 15 = 15 Marks)
16.a Given two sentences x1 = “He had pills”, x2 = “Fan is running”. [BL4 [CO6] [15]
Design a Word2Vec model for processing the similarity between ]
these two sentences. Assume that each word is represented by
one-hot vector.
OR
16.b Sentence classification is the task of automatically assigning [BL4 [CO5] [15]
categories to sentences based on their content. This has broad ]
applications like identifying spam emails, classifying customer
feedback, or determining the topic of a news article.
i. Construct a suitable algorithm for solving this particular
task.
ii. Justify the reasons for choosing such algorithm.
iii. Implement the algorithm using Python.

**********

You might also like