8/6/25, 8:29 PM Deep Learning - IIT Ropar - - Unit 4 - Week 1
(https://swayam.gov.in) (https://swayam.gov.in/nc_details/NPTEL)
harshpal5276@gmail.com
NPTEL (https://swayam.gov.in/explorer?ncCode=NPTEL) » Deep Learning - IIT Ropar (course)
Click to register
for Certification
exam
Week 1 : Assignment 1
(https://examform.nptel.ac.in/2025_10/exam_form/dashboard)
Assignment not submitted Due date: 2025-08-06, 23:59 IST.
If already 1) For an MP neuron with input size 3 and threshold condition 1 point
registered, click f (x) = 1 if x 1 + x 2 + x 3 > 1
to check your Which of the following input-output pairs are correct?
payment status
𝑥 = (1,0,0), 𝑦 = 0
𝑥 = (0,1,1), 𝑦 = 1
𝑥 = (0,0,0), 𝑦 = 0
Course
𝑥 = (1,1,0), 𝑦 = 1
outline
𝑥 = (1,1,1), 𝑦 = 0
About NPTEL
() Data for the following 3 questions (2, 3, 4)
How does an A smart home system uses three binary sensors:
NPTEL online x 1 : Door sensor (1 if open)
course work? : Motion detector (1 if movement)
x2
()
x3 : Window sensor (1 if open)
The MP neuron outputs 1 only when the total number of activated sensors exceeds
Week 0 ()
2. This output triggers an alert to the homeowner.
Week 1 () 2) What is the minimum number of input values that must be 1 for the MP neuron to output 1?
Biological Fill in the Textbox (type answer) ___________
Neuron (unit?
unit=17&lesson
=18)
1 point
https://onlinecourses.nptel.ac.in/noc25_cs106/unit?unit=17&assessment=323 1/8
8/6/25, 8:29 PM Deep Learning - IIT Ropar - - Unit 4 - Week 1
From Spring to 3) Which of the following sensor combinations would trigger the alert? 1 point
Winter of AI
(unit? 𝑥 = (1, 1, 0)
unit=17&lesson 𝑥 = (0, 1, 1)
=19)
𝑥 = (1, 1, 1)
The Deep 𝑥 = (1, 0, 0)
Revival (unit?
𝑥 = (0, 0, 1)
unit=17&lesson
=20)
4) For McCulloch-Pitts (MP) neuron model, which of the below thresholding functions 1 point
From Cats to describes the above scenario.
Convolutional
Neural
Networks (unit? 1 if x 1 + x 2 + x 3 > 2
f (x) = {
unit=17&lesson 0 Otherwise
=21)
1 if x 1 + x 2 + x 3 ≥ 2
Faster, higher, f (x) = {
0 Otherwise
stronger (unit?
unit=17&lesson
=22) 1 if x 1 + x 2 + x 3 > 1
f (x) = {
0 Otherwise
The Curious
None of the above
Case of
Sequences
(unit? 5) A team designs an MP neuron to detect when at least two out of three emergency 1 point
unit=17&lesson switches are turned on. You realize the current model fires only when all three are on.
=23) Which change will help the model fire (output 1) when at least two inputs are 1?
Beating
Change the threshold from 2 to 1
humans at their
own games Change the condition from “> 2” to “≥ 2”
(literally) (unit? Use weights instead of plain sums
unit=17&lesson
Set threshold to 3
=24)
Make all inputs always 1
The Madness
(2013-) (unit?
6) A fault-tolerant 7-input MP neuron must output 1 only if a majority (i.e., more than 50%) 1 point
unit=17&lesson
of inputs are ON. The neuron receives binary signals (0/1). Which of the following threshold values
=25)
will ensure that the neuron functions as a strict majority gate?
(Need for)
Sanity (unit? 3
unit=17&lesson 4
=26)
5
Motivation from 6
Biological
Neurons (unit?
7) Consider an MP neuron that takes 4 binary inputs with the following learning rule. 1 point
unit=17&lesson
=27) f (x) = {
1 if x 1 + x 2 + x 3 ≥ 1
0 Otherwise
McCulloch Pitts For how many input combinations does it output 1?
Neuron,
Thresholding
https://onlinecourses.nptel.ac.in/noc25_cs106/unit?unit=17&assessment=323 2/8
8/6/25, 8:29 PM Deep Learning - IIT Ropar - - Unit 4 - Week 1
Logic (unit? 8
unit=17&lesson
=28)
15
1
Perceptrons
(unit? 0
unit=17&lesson
8) A security system has 4 sensors (x 1 to x 4 ). An MP neuron-based controller triggers an alarm
=29)
when at least 3 sensors are active. How many unique input cases will lead to an alarm?
Error and Error
Surfaces (unit? Fill in the blank : ____________
unit=17&lesson
=30)
1 point
Perceptron
Learning 9) An MP neuron with 4 binary inputs gives output 1 for 11 different input combinations. 1 point
Algorithm (unit?
What is the likelihood of 𝜃?
unit=17&lesson
=31)
2
Proof of 3
Convergence
1
of Perceptron
Learning 4
Algorithm (unit?
unit=17&lesson 10) You're building a neural circuit using a single perceptron that needs to identify when 1 point
=32) both security switches are on. Which Boolean function would you implement to solve this?
Lecture
OR
Material for
Week 1 (unit? XOR
unit=17&lesson AND
=33)
XNOR
Week 1
Feedback 11) You’re designing a logic gate using a perceptron. You want the gate to output 1 only 1 point
Form:Deep when exactly one of the two inputs is 1.
Learning - IIT
Which function are you trying to implement? Can a single perceptron do it?
Ropar!! (unit?
unit=17&lesson AND, Yes
=35)
XOR, No
Quiz: Week 1 : OR, Yes
Assignment 1
(assessment?
NOR, Yes
name=323)
12) In a smart voting system, a gate opens only if exactly one person presses a button out 1 point
Week 2 () of two users (User A and User B).
Which logic function represents this? Can it be modeled by a perceptron?
Week 3 ()
XOR, Yes
Download XOR, No
Videos () AND, Yes
NOR, No
https://onlinecourses.nptel.ac.in/noc25_cs106/unit?unit=17&assessment=323 3/8
8/6/25, 8:29 PM Deep Learning - IIT Ropar - - Unit 4 - Week 1
13) A team is developing an AI system for identifying whether a tumor is malignant or 1 point
benign based on two numerical features extracted from medical images. They decide to use a
simple linear model for this binary decision task. They come across the perceptron algorithm.
Which of the following best justifies the use of the perceptron algorithm in this task?
To group similar tumors into clusters
To classify tumors as benign or malignant
To compute the shortest path from data to prediction
To solve a constrained optimization problem
14) You are designing a 2D binary classifier using the perceptron algorithm. After training 1 point
the model, you decide to visualize the decision boundary and the weight vector w⃗ . You notice that
the line dividing the data and the vector w⃗ make an angle. Which of the following best explains the
relationship between the weight vector w⃗ and the decision boundary?
The weight vector is always parallel to the decision boundary
The weight vector lies on the decision boundary
The weight vector is perpendicular to the decision boundary
The weight vector changes direction randomly with each epoch
Data for questions 15 and 16 :
Consider you are training a perceptron algorithm on the following data, where x1 and x2 are
features and y is a label.
15) Assume that the elements in 𝑤 are initialized to [ 0.0, 0.0 ] and the perceptron learning 1 point
algorithm is used to update the weight vector 𝑤. If the learning algorithm runs for long enough
iterations then
The algorithm never converges
The algorithm converges (i.e., no further weight updates) after some iterations
The classification error remains greater than zero.
The classification error becomes zero eventually
16) Assume the weights are initialized randomly in the range [-0.5, 0.5], then which of the 1 point
following statements are true?
https://onlinecourses.nptel.ac.in/noc25_cs106/unit?unit=17&assessment=323 4/8
8/6/25, 8:29 PM Deep Learning - IIT Ropar - - Unit 4 - Week 1
The perceptron may fail to converge due to random initialization
The perceptron will converge regardless of weight initialization
Convergence time may vary depending on the initial weights
Random initialization can help the model escape local minima
17) You have a trained perceptron model, whose final weights are 𝑤 = [1, 1] and 𝑏 = −1. 1 point
The learning rule of the trained perceptron model is
1 if w. x + b > 0
y = {
0 otherwise
Which of the following (x 1 , x 2 , y ) samples would be correctly classified by this model?
(1, 1, 1)
(0, 0, 1)
(1, 0, 0)
(0, 1, 0)
18) You're designing a voting system for a critical application with 7 redundant sensors. To 1 point
ensure fault tolerance, the system should activate only when all sensors agree. You use an MP
neuron to model this.
Which of the following threshold values will ensure true activation only when all 7 inputs are ON?
5
6
7
8
19) In a voting system with 7 voters, at least 5 voters must vote YES for the system to proceed.
You want to model this with an MP neuron.
What threshold value should you set?
Fill the blank____
1 point
20) Consider the following perceptron learning rule 1 point
T
1 if w x > 0
y = {
−1 otherwise
−1
Assume the above perceptron predicts 𝑦 = 1, for the weight vector 𝑤 = [ ] , and a point 𝑥 = (𝑎,
−1
𝑏). Which of the following must be true about the relation between 𝑎 and 𝑏?
𝑎+𝑏<0
𝑎+𝑏=0
𝑎+𝑏>0
𝑎𝑏 > 0
https://onlinecourses.nptel.ac.in/noc25_cs106/unit?unit=17&assessment=323 5/8
8/6/25, 8:29 PM Deep Learning - IIT Ropar - - Unit 4 - Week 1
21) −1 1 point
Consider a perceptron with w = [ ] and no bias classifies a 2D input vector using
−1
the rule: y = sign(w
T
. Which of the following points lie on the decision boundary?
x)
(2, 2)
(1, −1)
(3, −3)
(0.5, 0.5)
22) Given a perceptron with weight vector [2, 3, −1] and bias −2, what will be the output when the
input vector is [1, 0, −1]?
Fill the answer : _____________
1 point
Questions 23 to 25
Instructions for the following questions
Assume perceptron learning without bias and use the following update rules:
If the example is from class 𝑦 = 1 and 𝑤. 𝑥 < 0, then 𝑤 = 𝑤 + 𝑥
If the example is from class 𝑦 = 0 and 𝑤. 𝑥 ≥ 0, then 𝑤 = 𝑤 − 𝑥
All feature vectors have two dimensions. Pick the examples in the order mentioned in the table.
23) The perceptron is initialized with weight vector 𝑤 = (3,3). Consider the following training
dataset.
How many updates will the perceptron perform in one full pass over the data?
Fill the blank : ____________
1 point
24) The dataset is 1 point
https://onlinecourses.nptel.ac.in/noc25_cs106/unit?unit=17&assessment=323 6/8
8/6/25, 8:29 PM Deep Learning - IIT Ropar - - Unit 4 - Week 1
Assume perceptron starts with 𝑤 = (0, 0), and training continues until convergence.
Which of the following could be a valid final weight vector after convergence?
(1, 0)
(-1, -1)
(0, 1)
(1, 1)
25) The dataset is 1 point
Assume perceptron learning starts with 𝑤 = (0, 0). The algorithm continues until convergence.
What is the final weight vector after convergence?
(1,2)
(-2, -1)
(-1,-1)
None of these
26) Which of the following points best explains why adding a bias term in a perceptron 1 point
model helps in classification?
It shifts the decision boundary away from the origin
It rotates the decision boundary
It increases the number of epochs required for convergence
None of the above
27) Which of the following conditions must be true for the perceptron learning algorithm to 1 point
guarantee convergence?
The data must be linearly separable
The weight vector must be initialized to non-zero values
The number of data points must be finite
The algorithm must include a bias term
28) What does it mean for two sets P and N to be absolutely linearly separable in 1 point
ndimensional space?
There exists a hyperplane that correctly classifies all points from both sets
Every point from P lies on the same coordinate axis
https://onlinecourses.nptel.ac.in/noc25_cs106/unit?unit=17&assessment=323 7/8
8/6/25, 8:29 PM Deep Learning - IIT Ropar - - Unit 4 - Week 1
Each set must have an equal number of points
The points in P are closer to the origin than those in N
You may submit any number of times before the due date. The final submission will be considered
for grading.
Submit Answers
https://onlinecourses.nptel.ac.in/noc25_cs106/unit?unit=17&assessment=323 8/8
