Worksheet: Understanding 3.

The process of adjusting

Neural Networks connection weights in a neural
network based on training data
Class: XII is governed by:
A. Signal Propagation
B. Learning Rules
I. Multiple Choice Questions
(MCQ) (20 Questions) C. Activation Functions
Instructions: Choose the most D. Input Layer
appropriate option.
4. Which type of neural network is
1. Which component of a neural
network is primarily most effective for sequential
responsible for introducing non- data like text or speech due to its
linearity to the model, enabling use of feedback loops?
it to learn complex patterns? A. Feedforward Network
A. Weights B. Convolutional Neural Network
B. Bias (CNN)

C. Activation Functions C. Recurrent Neural Network

(RNN)
D. Input Layer
D. Deep Neural Network (DNN)
2. A neural network with two or
more hidden layers is 5. In the calculation Σwixi+bias,
specifically known as a: what does the term Σwixi
represent?
A. Feedforward Neural Network
A. The final output of the neuron
B. Convolutional Neural Network
B. The activation function's result
C. Recurrent Neural Network
C. The weighted sum of the inputs
D. Deep Neural Network
D. The error/loss value
6. Which application is not D. Bias Layer
explicitly mentioned in the
source material as a primary 9. What is the significance of the
use case for Neural Networks?
weight associated with a
A. Facial Recognition connection between neurons?
B. Machine Translation A. It determines the activation
function type.
C. Stock Market Prediction
B. It represents the strength or
D. Quantum Computing
importance of that
Simulation
feature/connection.
7. The main advantage of neural C. It specifies the number of
networks over traditional
hidden layers.
programming, as per the text, is
their ability to: D. It always remains constant
during training.
A. Process information faster than
the human brain. 10. Backpropagation is primarily
used to:
B. Be easily understood and
debugged by programmers. A. Compute the initial weights of
the network.
C. Extract data features
automatically without B. Compute the final output
programmer input. (forward pass).
D. Work exclusively with C. Adjust the weights and biases
sequential data. to minimize the network's
error/loss.
8. Which layer in a neural network
is responsible for generating the D. Determine the number of
final predictions or outputs? neurons in a layer.
A. Input Layer 11. What is the role of the Bias
term in a neuron's calculation?
B. Hidden Layers
C. Output Layer
 A. It scales the output of the D. The movement of planets in
activation function. the solar system.
B. It connects different layers. 14. The simplest neural network
with a single layer of input
C. It allows the network to shift nodes fully connected to a layer
the activation function of output nodes is called a:
horizontally and account for
inherent data bias. A. Feedforward Neural Network
(FFNN)
D. It calculates the weighted sum
of inputs. B. Recurrent Neural Network
(RNN)
12.A Convolutional Neural
Network (CNN) is particularly C. Perceptron (Standard Neural
effective for which type of Network)
application?
D. Generative Adversarial
A. Financial time-series Network (GAN)
prediction
15. What happens when the
B. Machine translation output of a node exceeds a
specified threshold value?
C. Image Recognition and
Computer Vision tasks A. The weight of the connection
is reset.
D. Basic binary classification
(spam detection) B. The node is activated, and its
output is passed to the next layer.
13. Which concept is loosely
mimicked by an Artificial C. The network begins the
Neural Network (ANN)? backpropagation process.
A. The flow of current in a D. The bias term is adjusted.
circuit.
16.Which ethical concern is
B. How the human brain explicitly emphasized regarding
processes information. the use of Neural Networks?
C. The structure of a binary A. Over-reliance on
search tree. computational power.
 B. Complexity of network 19.In a Feedforward Neural
architecture. Network (FFNN), how does
information flow?
C. Data privacy and algorithmic
bias. A. In a loop, from output back to
input.
D. Lack of real-world
applications. B. In one direction, from the input
layer to the output layer.
17. In the example calculation for
the hidden layer (Case I), what C. Simultaneously across all
leads to the output being 0 neurons regardless of layer.
(inactive)?
D. Primarily between nodes in the
A. The bias value was too low. same hidden layer.
B. The input values were too 20.What is the fundamental
small. building block of a neural
network, also known as a node?
C. The calculated output (2.1) was
less than the threshold (3.0). A. Layer
D. The weights were all less than B. Connection
18. What is the role of the C. Weight
Discriminator network in a
Generative Adversarial D. Neuron
Network (GAN)?
A. To create new data instances. II. Assertion and Reason (10
B. To adjust the weights of the Questions)
Generator. Instructions: In the following
C. To evaluate the generated data questions, a statement of
for authenticity (real/fake). Assertion (A) is followed by a
statement of Reason (R). Choose
D. To propagate the input data the correct option.
forward.
 Options Reason (R): Activation functions
determine a neuron's output based
Both A and R are true, on its weighted inputs and are like
A and R is the correct decision-makers for each neuron.
explanation of A.
4. Assertion (A): Generative
Both A and R are true, Adversarial Networks (GANs) are
B but R is not the correct used for unsupervised learning.
explanation of A.
Reason (R): GANs consist of two
C A is true, but R is false. networks, a generator that creates
new data and a discriminator that
D A is false, but R is true. evaluates its authenticity.

1. Assertion (A): Convolutional 5. Assertion (A): The learning rule

Neural Networks (CNNs) are specifies how the weights and
dominant in computer vision tasks biases are adjusted during
like image recognition. training.

Reason (R): CNNs incorporate a Reason (R): Backpropagation is a

three-dimensional arrangement of common learning algorithm that
neurons, which is highly effective computes gradients and updates
for processing visual data. weights to minimize the network's
error.
2. Assertion (A): The process of
training deep neural networks is 6. Assertion (A): Feedforward
called Deep Learning. Neural Networks (FFNNs) are
also known as Multi-Layer
Reason (R): The term "deep" Perceptrons (MLPs).
refers to the number of hidden
layers (depth) in a neural network. Reason (R): FFNNs are used in
tasks like image recognition,
3. Assertion (A): A neuron is NLP, and regression.
activated, and its output is passed
to the next layer only if the output 7. Assertion (A): The Bias terms
of the node exceeds a specified allow the network to shift the
threshold value. 25 activation function horizontally.
 Reason (R): Weights represent between the Input Layer and
the importance of a feature in the Output Layer.
predicting the final output.
8. Assertion (A): Neural Networks
have significantly enhanced 2. Define Backpropagation in one
efficiency by automating tasks sentence, as described in the
and optimizing resource provided text.
allocation in various industries.
Reason (R): One ethical concern 3. Name two specific real-world
related to NNs is potential job applications of Neural Networks
displacement. mentioned in the text, one for
Image Recognition and one for
9. Assertion (A): In the forward Natural Language Processing.
pass (Forward Propagation), the
input data flows through the
layers, and the predicted output is 4. What is the difference between
compared to the actual target. a basic neural network and a
Deep Neural Network in terms
Reason (R): This comparison of the number of layers?
results in an error (loss) which is
used in the backward pass.
5. What is the analogy used to
10. Assertion (A): A basic neural introduce the concept of Neural
network must have at least three Networks to students?
layers: input, hidden, and output.
Reason (R): A Perceptron IV. Long Answer Questions (2
(Standard Neural Network) only Questions)
has an input and an output layer. 40
1. Explain the working of a single
neuron (node) in a neural
III. Very Short Answer network using the mathematical
Questions (5 Questions) concept Σwixi+bias. Describe
the sequence of steps from input
1. In the context of the structure of
to output, including the roles of
a neural network, differentiate
 weights, bias, and the activation o Question: Which of the three
function/threshold. main types of neural networks
(FFNN, CNN, or RNN) is the
2. Discuss the societal impact of most appropriate architecture
Neural Networks, covering both for this task? Explain why this
the positive contributions and type is better suited than the others
the key ethical concerns that based on its characteristics
need careful consideration. described in the text.
Provide at least two points for
each side.
3. Scenario: The final output of a
V. Competency Based neuron is calculated as 4.5. The
Questions (5 Questions) required output for the target
category is 1.0.
1. Scenario: A company is
developing an advanced spam
email filter. They have a massive o Question (a): What is the term for
dataset of emails labeled as 'spam' the difference between the
or 'not spam'. calculated output (4.5) and the
target output (1.0)?

o Question: Which type of basic

neural network, described in the o Question (b): Which key
text, would be most suitable for propagation function/learning
this binary classification task? algorithm described in the text is
Justify your answer. employed to use this difference to
improve the network's future
predictions? Briefly explain what
2. Scenario: A machine learning it adjusts.
team wants to build a system that 4. Scenario: You are trying to build
can automatically describe the a neural network to decide on a
actions happening in a continuous loan application. The input
video feed (sequential data), for features are: Applicant Age (x1
example, "The car is turning left." =40, weight w1=0.2), Income (x2
=50000, weight w2=0.8), and
 Credit Score (x3=750, weight w3 Explain the Diagram
=1.5). The neuron's bias is −2000.

o Question: Calculate the weighted

sum (Σwixi+bias). Show your full
working. If the activation
threshold is 7000, will the neuron
activate (Output=1) or remain
inactive (Output=0)?

5. Scenario: A large e-commerce

platform uses a neural network to
recommend products to users
based on their past purchases and
browsing history.

o Question: Explain the ethical

concern of algorithmic bias in the
context of this recommendation
system. Suggest one potential way
to mitigate this bias, using
information from the text.