Artificial Neural Networks (ANN) - Insem Answer Key

Rudra
March 2026

1. February 2023 Paper

Q1
a) What is a neural network activation function? State its Types? [5] An activation function
(also known as a transfer function) is a mathematical model that determines the output of a neuron
based on the weighted sum of its inputs and a bias[cite: 189, 2901]. It introduces non-linearity into
the network, allowing it to learn complex patterns that a simple linear model cannot[cite: 193, 2914].
Without non-linearity, no matter how many layers a neural network has, it would behave like a single-
layer perceptron[cite: 193].
Types of Activation Functions:
• Identity / Linear Function: The output is directly proportional to the input (f (x) = x).
It is mainly used in the output layer for regression tasks where a range of continuous values is
needed[cite: 198, 200].
• Binary Sigmoid Function: It maps any real-valued input into a range between 0 and 1. It is
commonly used for binary classification where the output represents a probability[cite: 206, 2906].
• Bipolar Sigmoid Function: Similar to the binary sigmoid but produces outputs in the range of
-1 to +1. This helps in mapping inputs that are strongly negative or positive[cite: 3375, 3376].
• Hyperbolic Tangent (Tanh): It is a zero-centered function with a range of -1 to 1. It is
generally preferred over the sigmoid function because it models inputs more effectively by being
zero-centered[cite: 214, 2917].
• ReLU (Rectified Linear Unit): It returns 0 for negative inputs and the input itself for positive
values (f (x) = max(0, x)). It is the most used activation function today due to its computational
efficiency[cite: 216, 219].

• Leaky ReLU: A variation of ReLU that allows a small, non-zero gradient when the input is
negative. This prevents the ”dying ReLU” problem where neurons stop learning entirely[cite:
2919, 2422].
• Softmax Function: It converts a vector of raw scores into probabilities that sum to 1. It is almost
exclusively used in the output layer of multi-class classification networks[cite: 2422, 2921].

Reference: [Link] (1-18 to 1-22), ANN Module 1 Lecture [Link] (8-9)

b) Explain architecture of Artificial Neural Network with a neat diagram. [5] The archi-
tecture of an Artificial Neural Network refers to the arrangement of neurons in layers and the pattern of
connections between them[cite: 105, 2583]. It typically consists of three primary types of layers:

1. Input Layer: This is the first layer of the network that receives raw data from the external
environment[cite: 70, 2589]. It does not perform any computation; it simply passes the features
(like image pixels or numerical data) to the next layer[cite: 2453].
2. Hidden Layer(s): These layers reside between the input and output layers and are responsible
for the actual processing[cite: 70, 2590]. They perform complex calculations and extract patterns
from the data through weighted sums and activation functions[cite: 2388].

1
 3. Output Layer: This is the final layer that produces the result of the network’s processing[cite:
70, 2591]. It provides the final prediction, such as a category label in classification or a numerical
value in regression[cite: 2457].

Key Components:
• Weights: Represent the strength of the connection between two neurons. During training, these
are adjusted to minimize error[cite: 72, 314].

• Bias: A trainable constant added to the weighted sum, allowing the activation function to shift
for better data fitting[cite: 118, 3001].
• Summation Function: Computes the net input by adding all weighted inputs and the bias[cite:
3167, 3180].

[Insert Diagram Here – Source: [Link], Page No: 1-2 (Fig 1.1.1)]
Reference: [Link] (1-2, 1-8), ANN [Link] (4, 10)

c) Explain Mc-Culloch & Pitts model with an example. [5] The McCulloch-Pitts (M-P) model,
proposed in 1943, was the first mathematical model of a biological neuron[cite: 221, 3005]. It represents
a neuron as a simple logic threshold gate that processes binary inputs to produce a binary output[cite:
224, 3015].
Key Features:

• Binary Inputs/Outputs: Inputs and outputs are strictly {0, 1}. A ’1’ represents an active state,
and ’0’ represents a silent state[cite: 226, 3009].
• Weights: All excitatory inputs have identical, fixed positive weights, while inhibitory inputs have
absolute veto power[cite: 240, 241].

• Threshold (θ): The neuron fires (output = 1) if the sum of excitatory inputs reaches or exceeds
the threshold θ, provided no inhibitory input is present[cite: 242, 3023].
Example: AND Function To implement a 2-input AND function, both inputs x1 and x2 must be
1 for the output y to be 1. We set the weights w1 = w2 = 1 and the threshold θ = 2[cite: 3024].
• If (x1 , x2 ) = (1, 1), sum = 2 ≥ θ, output = 1.

• For any other input, sum < 2, output = 0[cite: 3024].

Reference: [Link] (1-23 to 1-24), ANN Module 1 Lecture [Link] (11-13)

Q2
a) What are the main differences among the three models of artificial neuron, namely,
McCulloch-Pitts, Perceptron and Adaline? [5]

Feature McCulloch-Pitts Perceptron Adaline

Year/Creator 1943 (McCulloch & 1958 (Rosenblatt) 1960 (Widrow & Hoff)
Pitts) [cite: 221] [cite: 715] [cite: 719]
Inputs Binary (0, 1) [cite: Real or Boolean [cite: Real-valued [cite: 303]
239] 258]
Weights Fixed / Analytically Learnable / Ad- Learnable / Ad-
determined [cite: 247] justable [cite: 263] justable [cite: 3093]
Learning Rule No learning (Hard- Perceptron Learning Delta Rule / LMS Rule
wired) [cite: 247] Rule [cite: 268] [cite: 309]
Activation Step Function [cite: Step or Signum [cite: Bipolar / Linear Sum
238] 122] [cite: 3091]
Error Minimize Not applicable [cite: Binary classification Mean Square Error
247] error [cite: 265] (MSE) [cite: 3092]

Reference: [Link] (1-23, 1-25, 1-29), ANN Module 1 Lecture [Link] (11, 15)

2
b) Explain the structure and working of Biological Neural Network? [5] A Biological Neural
Network (BNN) is the complex web of neurons in the human brain responsible for learning and decision-
making[cite: 2556, 2557].
Structure of a Biological Neuron:

• Dendrites: These are tree-like branches that act as the input unit, receiving electrical signals from
other neurons[cite: 53, 2559].
• Soma (Cell Body): The central processing unit that integrates all incoming signals and decides
whether to fire[cite: 57, 2560].

• Axon: A long cable-like transmission unit that carries the electrical impulse away from the soma
toward other neurons[cite: 61, 2561].
• Synapse: The junction or communication point where signals are passed from one neuron to
another using neurotransmitters[cite: 54, 2562].
Working Mechanism: Signals enter through dendrites and accumulate in the soma[cite: 2565,
2566]. If the total electrochemical potential exceeds a specific threshold, an ”action potential” or spike is
generated[cite: 2567, 3224]. This spike travels down the axon to the synapse, where it triggers the release
of chemicals to stimulate the next neuron in the circuit[cite: 2568, 2569]. Reference: [Link] (1-4),
ANN [Link] (8), ANN Module 1 Lecture [Link] (1-2)

c) Differentiate between Biological Neural Network and Artificial Neural Network. [5]

Feature Biological Neural Network Artificial Neural Network

Component Soma, Dendrites, Axons, Synapses Units, Connections, Weights, Bias [cite:
[cite: 65, 2558] 65, 2586]
Processing Asynchronous and Parallel [cite: 2848] Synchronous and Parallel [cite: 2848]
Speed Slow (milliseconds) [cite: 104, 3277] Fast (nanoseconds/microseconds) [cite:
104, 3277]
Learning Continuous, using unknown mecha- Epoch-based, using Gradient Descent
nisms [cite: 3300] [cite: 3300]
Storage Memory is spatially distributed [cite: Memory is stored in numeric weights
346] [cite: 2988]
Fault Tolerance Highly fault-tolerant (plasticity) [cite: Limited; depends on redundancy [cite:
2855] 2856]

Reference: [Link] (1-5, 1-8), ANN Module 1 Lecture [Link] (4-5)

Q3
a) Explain Perceptron Learning Algorithm with an example. [5] The Perceptron Learning
Algorithm is a supervised learning method for training a single-layer network to classify linearly separable
data[cite: 249, 3048].
Steps of the Algorithm:
1. Initialization: Assign random small values to weights (wi ) and bias (b)[cite: 263, 3036].
2. Activation: Apply an input vector x from the training set and compute the actual output y[cite:
125, 3049].
3. Weight Update: Compare y with desired output d. If y ̸= d, update weights using: ∆wi =
η(d − y)xi , where η is the learning rate[cite: 266, 3059].
4. Iteration: Repeat for all training pairs until no errors occur (convergence)[cite: 269, 933].

Example: OR Function For inputs (1, 0), desired d = 1. If initial weights give y = 0, the error is
1. The algorithm increases the weights so that the neuron is more likely to fire for this input in the next
iteration[cite: 3060, 3118]. Reference: [Link] (1-11, 1-27), ANN Module 1 Lecture [Link]
(15-16)

3
b) Explain the architecture of Multilayered neural network. [5] A Multilayered Neural Net-
work (also called Multi-Layer Perceptron or MLP) is a feed-forward network containing one or more
hidden layers between input and output[cite: 271, 144].
Architecture Components:

• Input Layer: A set of source nodes that distribute input values to the first hidden layer without
performing calculations[cite: 971, 275].
• Hidden Layers: These are the ”computational engine” of the network[cite: 284]. Neurons in
these layers use non-linear activation functions (like Sigmoid or ReLU) to learn non-linear relation-
ships[cite: 286, 142].

• Output Layer: The final layer that performs the specific task, such as classification into N classes
or predicting a continuous value[cite: 283].
• Interconnections: Every neuron in one layer is typically connected to all neurons in the subse-
quent layer, but no connections exist within the same layer[cite: 974, 607].

Reference: [Link] (1-12, 1-28), ANN [Link] (10-11)

c) Write and explain the steps of Back Propagation Learning algorithm. [5] Back Propaga-
tion is a systematic gradient descent method for training multilayer networks by minimizing the Mean
Square Error (MSE)[cite: 540, 541].
Steps of the Algorithm:
1. Initialization: Initialize weights and biases to small random values to avoid saturation[cite: 559,
3143].
2. Forward Pass: Apply an input pattern and propagate it through hidden layers to compute the
network’s final output[cite: 565, 3144].
3. Error Computation: Calculate the difference between the actual output and the target value
(loss function)[cite: 561, 3145].

4. Backward Pass: Propagate the error signal backward from the output layer to hidden layers
using the chain rule to determine each weight’s contribution to the error[cite: 566, 3146].
5. Weight Update: Adjust weights in the direction that reduces error: wnew = wold −η(∂E/∂w)[cite:
562, 2996].

Reference: [Link] (2-20 to 2-22), ANN Module 1 Lecture [Link] (23-26)

Q4
a) Draw the architecture of Back Propagation Network and explain in detail. [5] The
Back Propagation Network (BPN) architecture is typically a multilayer feed-forward structure where
information flows forward and error correction flows backward[cite: 564, 3159].
Detailed Structure: It consists of an input layer, at least one hidden layer, and an output layer[cite:
3157]. Each unit (except input) computes a weighted sum of its inputs, adds a bias, and applies a
differentiable activation function like the Logistic Sigmoid[cite: 546, 548]. The connections are strictly
feed-forward during the operation phase[cite: 594]. During the learning phase, the error computed at
the output layer is used to adjust weights across all layers in reverse order[cite: 569, 570].
[Insert Diagram Here – Source: [Link], Page No: 2-20 (Fig 2.7.1)]
Reference: [Link] (2-20 to 2-22), ANN Module 1 Lecture [Link] (26)

b) Differentiate between Feed forward and Feedback neural network. [5]

Feature Feedforward Neural Network Feedback (Recurrent) NN

Data Flow Unidirectional (Input → Hidden → Bidirectional; contains loops [cite: 608,
Output) [cite: 591, 2398] 3320]

4
 Memory No memory; output depends only on Has internal memory of previous inputs
current input [cite: 64, 2410] [cite: 180, 2410]
Complexity Simple architecture; easier to train Complicated; prone to vanishing gradi-
[cite: 592] ents [cite: 609, 2412]
Stability Highly stable Can be unstable until reaching equilib-
rium [cite: 3321]
State Static mapping [cite: 359] Dynamic system [cite: 360, 609]
Applications Pattern recognition, simple classifica- Time-series prediction, NLP [cite: 182,
tion [cite: 595] 1109]

Reference: [Link] (2-23 to 2-25), Module-II [Link] (22-23)

c) What is Error Correction and how to minimize these errors? [5] Error Correction Learning
is a supervised learning paradigm where ANN parameters are adjusted based on the difference between
the desired and actual output[cite: 449, 3991].
Minimization Process:
• Error Signal (e): Defined as e = d − y (Desired - Actual)[cite: 453, 3334].

• CostPFunction: Errors are squared and averaged to form the Mean Square Error (MSE) J =
E[ 21 e2 ][cite: 455, 332].
• Gradient Descent: The algorithm calculates the gradient of the cost function with respect to
weights and moves weights in the opposite direction (downhill) to find the local minimum[cite: 722,
542].

• LMS / Widrow-Hoff Rule: A specific rule that updates weights incrementally after each training
sample to minimize MSE[cite: 309, 3092].
Reference: [Link] (2-11 to 2-13), ANN Module 1 Lecture [Link] (22)

5
2. March 2024 Paper
Q1
a) What is the role of activation function in neural network? Explain bipolar Sigmoid func-
tion in detail. [5] The primary role of an activation function is to introduce non-linearity, enabling
the network to learn complex mappings and limiting the output signal range (normalization)[cite: 190,
193]. It also facilitates backpropagation as its derivative guides weight updates[cite: 191, 3383].
Bipolar Sigmoid Function: The bipolar sigmoid function maps input values into the range
(−1, 1)[cite: 3375]. Its mathematical form is f (x) = 1+e2−x − 1[cite: 3378]. It is strictly increasing
and bounded, meaning higher inputs result in higher outputs up to a limit of +1[cite: 3376]. Unlike
the binary sigmoid, it is symmetric around the origin, which often leads to faster convergence during
training as the data becomes ”zero-centered”[cite: 216, 2917]. Reference: [Link] (1-18, 1-21),
ANN Module 1 Lecture [Link] (9)

b) Why is ReLU the most commonly used Activation Function? [5] ReLU (Rectified Linear
Unit) is the industry standard due to several key advantages:
• Computational Efficiency: It involves simple thresholding (max(0, x)) rather than expensive
exponential calculations required by Sigmoid or Tanh[cite: 219, 2918].
• Sparse Activation: Since it outputs zero for all negative inputs, it effectively ”turns off” non-
contributing neurons, making the network lighter and more efficient[cite: 219].
• Reduced Vanishing Gradient: For positive inputs, the gradient is constant (1), which prevents
the gradient from shrinking to zero in deep networks[cite: 220, 2924].
• Faster Convergence: Empirical studies show that ReLU-based networks converge significantly
faster than those using saturating functions like Tanh[cite: 220].
Reference: [Link] (1-22 to 1-23), ANN Module 1 Lecture [Link] (9)

c) Explain architecture of Artificial Neural Network with a neat diagram. [5] Refer Answer:
Paper February 2023 – Question Q1(b)

Q2
a) Draw the structure of the biological neuron and explain working of the same in brief.
[5] Refer Answer: Paper February 2023 – Question Q2(b)

b) Write an algorithm of ADALINE and focus on its upper bound with largest Eigen Value
of its correlation matrix. [10] ADALINE (Adaptive Linear Neuron) uses the Delta rule to minimize
Mean Square Error[cite: 299, 3092].
ADALINE Algorithm Steps:
1. Initialize: Set weights, bias, and learning rate α to 0 or small random values[cite: 3096].
2. Loop: For each training pattern (x, t):
P
3. Net Input: Compute yin = b + xi wi [cite: 3098].
4. Activation: Apply identity or bipolar step for final output y[cite: 3099].
5. Weight Update: If error exists, update using: wnew = wold + α(t − yin )x and bnew = bold + α(t −
yin )[cite: 3099].
6. Convergence: Stop when weights stop changing or MSE reaches a minimum tolerance[cite: 3100].
Upper Bound and Eigen Values (Source: Internet): In ADALINE training, the stability and
speed of convergence are governed by the learning rate α. For the Least Mean Square (LMS) algorithm
2
to converge, the learning rate must satisfy: 0 < α < λmax , where λmax is the largest Eigenvalue of the
T
input correlation matrix R = E[xx ]. If α exceeds this upper bound, the weight updates will overshoot
the minimum and the network becomes unstable. A smaller α ensures stability but slows down learning.
Reference: [Link] (1-29 to 1-31), GeeksforGeeks (Internet Source)

6
Q3
a) What is Error Correction and how to minimize these errors? [5] Refer Answer: Paper
February 2023 – Question Q4(c)

b) Explain the architecture of Multilayered neural network. [5] Refer Answer: Paper February
2023 – Question Q3(b)

c) Define learning and memory. Explain learning algorithms in details. [5] Learning:
In ANN, learning is the iterative process of modifying weights and biases based on data to improve
performance on a specific task[cite: 311, 446]. Memory: Memory refers to the distributed information
contained within the adjusted synaptic weights of the network[cite: 24, 703].
Types of Learning Algorithms:
• Supervised Learning: Requires a teacher to provide target outputs. Weights are adjusted based
on the error between actual and target outputs (e.g., Backpropagation)[cite: 488, 3337].
• Unsupervised Learning: The network discovers hidden patterns in input data without any
external target (e.g., Clustering using Hebbian learning)[cite: 3342, 3347].
• Reinforcement Learning: Learning based on an evaluative reward/penalty signal from the
environment rather than explicit instructions[cite: 3351, 3354].

Reference: [Link] (2-1 to 2-2), Module-II [Link] (1, 13-15)

Q4
a) What is the difference between Forward propagation and Backward Propagation in
Neural Networks? [5]

Feature Forward Propagation Backward Propagation

Direction Input → Hidden → Output [cite: 565, Output → Hidden → Input [cite: 566,
3143] 3136]
Purpose To generate a prediction or result [cite: To calculate error and update weights
3144] [cite: 3146]
Operation Weighted sum and activation [cite: 567, Partial derivatives and chain rule [cite:
3128] 566, 3147]
Input Raw data / features [cite: 3128] Error signal / loss [cite: 179, 3139]
Dependency Independent of error [cite: 3159] Highly dependent on forward pass re-
sult [cite: 3145]
Timing Occurs first in every iteration [cite: 564] Occurs after the error is calculated [cite:
564]

Reference: [Link] (2-22), ANN Module 1 Lecture [Link] (24-25)

b) Explain the different types of Gradient Descent in detail. [5] Gradient Descent is an
optimization algorithm used to minimize the cost function[cite: 722].
• Batch Gradient Descent: Calculates the error for the entire dataset before updating weights.
It provides a stable error gradient but is very slow for large datasets[cite: 590, 3151].
• Stochastic Gradient Descent (SGD): Updates weights after each individual training example.
It is much faster and can avoid local minima but results in a noisy convergence path[cite: 458, 460].
• Mini-Batch Gradient Descent: A compromise that updates weights after a small batch of
samples (e.g., 32 or 64). It combines the stability of batch GD with the efficiency of SGD[cite: 590,
3151].
Reference: [Link] (2-13), ANN Module 1 Lecture [Link] (25)

7
c) Write down Perceptron Learning Algorithm for OR function along with calculation of
each input vector. [5] Let initial weights w1 = 0, w2 = 0, b = 0 and η = 1. Threshold θ = 0. Inputs:
X1 (0, 0), X2 (0, 1), X3 (1, 0), X4 (1, 1). Targets: T (0, 1, 1, 1).
Calculations: 1. Input (0,0): yin = 0, y = 0. y = T , no change. 2. Input (0,1): yin = 0,
y = 0. y ̸= T . Update: ∆w2 = 1(1 − 0)1 = 1, ∆b = 1. New w = [0, 1], b = 1. 3. Input (1,0):
yin = 1(0) + 0(1) + 1 = 1, y = 1. y = T , no change. 4. Input (1,1): yin = 1(1) + 1(1) + 1 = 3, y = 1.
y = T , no change. Final weights: w = [0, 1], b = 1 correctly classifies OR function[cite: 269, 3118].
Reference: [Link] (1-27), ANN Module 1 Lecture [Link] (15-16, 20)

8
3. March 2025 Paper
Q1
a) What is neural network topology? Explain any three topologies in detail. [6] Network
topology is the specific arrangement of nodes (neurons) and the connecting lines (weights) in a neural
network[cite: 105, 3033].
Topologies:
• Single-Layer Feedforward: The simplest topology where the input layer is directly connected
to the output layer with a single set of weights[cite: 3311, 3312]. There are no hidden layers, and
signals flow in one direction only[cite: 827, 2875].
• Multilayer Feedforward (MLP): Contains one or more hidden layers between the input and
output[cite: 3315, 3316]. It can learn complex, non-linear relationships and is the standard for
most classification tasks[cite: 2386, 2392].

• Recurrent Neural Network (RNN): A dynamic topology containing feedback loops where
information can flow backward[cite: 180, 1101]. This creates an internal memory, making it ideal
for sequential data like speech or time-series[cite: 1097, 2411].
Reference: [Link] (1-8, 1-17), ANN Module 1 Lecture [Link] (5-6)

b) Explain briefly McCulloch Pitt’s (MP) artificial neuron model. Give its limitations. [5]
Refer Architecture: Paper February 2023 – Question Q1(c).
Limitations of M-P Model:
• No Learning: Weights and thresholds are hard-coded and cannot be adjusted automatically from
data[cite: 247, 3028].
• Strictly Binary: It cannot handle real-valued inputs or continuous outputs, limiting its use to
basic logic[cite: 714, 244].

• Fixed Threshold: The threshold value is static and must be determined analytically for every
problem[cite: 247, 3028].
Reference: [Link] (1-23 to 1-25), ANN Module 1 Lecture [Link] (14)

c) Distinguish between Biological Neural Network and Artificial Neural Networks. [4]
Refer Answer: Paper February 2023 – Question Q2(c)

Q2
a) Discuss briefly the structure and function of a biological neuron. [7] Refer Answer: Paper
February 2023 – Question Q2(b). (Additionally, for 7 marks, add detail: The axon hillock is the most
excitable part of the neuron where the spike initiation occurs[cite: 3212, 3214]. Synaptic boutons release
neurotransmitters like acetylcholine to bridge the synaptic cleft chemically[cite: 63, 3218].) Reference:
[Link] (1-4), ANN Module 1 Lecture [Link] (2)

b) Obtain the output of the neuron Y for the network shown in figure using activation
function as Binary Sigmoidal and Bipolar Sigmoidal. [8] (Assuming a standard problem with
Inputs
P X = [0.5, 0.2], Weights W = [0.4, 0.1], Bias b = 0.3) 1. Calculate Net Input (S): S =
(xi wi ) + b = (0.5 × 0.4) + (0.2 × 0.1) + 0.3 = 0.2 + 0.02 + 0.3 = 0.52[cite: 3098]. 2. Binary Sigmoid
1 1
Output: f (S) = 1+e−0.52 ≈ 1+0.594 ≈ 0.627[cite: 3374]. 3. Bipolar Sigmoid Output: f (S) =
2
1+e−0.52 −1 ≈ 2(0.627)−1 ≈ 1.254−1 ≈ 0.254[cite: 3378]. Reference: ANN Module 1 Lecture [Link]
(9), [Link] (1-31)

9
Q3
a) Write and explain Hebbian learning Algorithm. [6] Proposed by Donald Hebb in 1949, it
states that ”neurons that fire together, wire together”[cite: 407, 3035].
Steps:
1. Initialize: Set all weights to zero or small random values[cite: 418, 3043].
2. Activate: Apply input x and compute output y[cite: 419, 3049].
3. Update: Modify weights using ∆wij = α · xi · yj , where α is the learning rate[cite: 413, 3047].

4. Iterate: Repeat for all patterns. Over time, connections between frequently co-active neurons
strengthen[cite: 408, 3042].
Reference: [Link] (2-7), ANN Module 1 Lecture [Link] (15)

b) What is error correction learning? Explain in detail with diagram. [5] Refer Answer:
Paper February 2023 – Question Q4(c).

c) Differentiate between Feed Forward and Feedback neural network. [4] Refer Answer:
Paper February 2023 – Question Q4(b).

Q4
a) Draw the architecture of multilayer feed forward networks. Explain input layer, hidden
layer & output layer computations in multilayer feed forward networks. [7] Refer Answer:
Paper February 2023 – Question Q3(b).
Layer Computations:
• Input Layer: Passes raw values xi without change (yi = xi )[cite: 2589].
• Hidden Layer: Computes hj = σ( wij xi + bj ), where σ is non-linear[cite: 2388].
P

• Output Layer: Computes Ok = σ( wjk hj + bk ) to produce final result[cite: 281, 282].

P

Reference: [Link] (1-28), ANN [Link] (11)

b) Explain perceptron learning algorithm and implement OR function using Perceptron

network. [8] Refer Answer: Paper March 2024 – Question Q4(c). (Additionally, mention the Per-
ceptron Convergence Theorem which states that the algorithm will always find a solution if the data is
linearly separable[cite: 933, 127].) Reference: [Link] (1-11, 1-27)

10