About this previous year paper
SECTION A – Short Answer Type Questions (2 Marks each)
(a) Discuss Model Representation of Artificial Neuron.
An artificial neuron is a mathematical model inspired by a biological neuron. It processes input signals to produce an output based on weights and activation function.
Mathematical Model:
y=f(∑i=1nwixi+b)y = f\left(\sum_{i=1}^{n} w_i x_i + b\right)y=f(i=1∑nwixi+b)
where:
xix_ixi: Input signals
wiw_iwi: Weights
bbb: Bias
fff: Activation function (e.g., sigmoid, ReLU)
yyy: Output
Explanation:
Inputs are multiplied by weights and summed with bias.
The activation function introduces non-linearity.
The neuron "fires" (produces output) if the total input exceeds a threshold.
Example: Used in perceptrons and neural networks for pattern recognition.
(b) Differentiate Between Gradient Descent and Stochastic Gradient Descent.
| Feature | Gradient Descent (GD) | Stochastic Gradient Descent (SGD) |
|---|---|---|
| Data Processing | Uses the entire dataset for one update. | Updates parameters after each sample. |
| Speed | Slower for large datasets. | Faster convergence with noise. |
| Accuracy | More stable and precise. | May fluctuate around minimum. |
| Memory Requirement | High (needs all samples). | Low (one sample at a time). |
Conclusion:
SGD is preferred for large datasets and online learning, while GD suits small or static datasets.
SECTION B – Long Answer Type Questions (10 Marks each)
(a) Explain Supervised and Unsupervised Learning Techniques.
1. Supervised Learning:
Input data is labeled, meaning the correct output is known.
The model learns a mapping function f(x)→yf(x) → yf(x)→y.
Examples:
Regression: Predicting house prices. Classification: Spam email detection.
Algorithms: Linear Regression, Decision Trees, SVM, Neural Networks.
2. Unsupervised Learning:
No labeled outputs; the model finds hidden patterns in data.
Examples:
Clustering: Customer segmentation using K-Means.
Association: Market basket analysis using Apriori algorithm.
Key Difference:
Supervised learning focuses on prediction, while unsupervised focuses on pattern discovery.
(b) Explain Various Types of Activation Functions with Examples.
Activation functions introduce non-linearity to neural networks, enabling complex pattern learning.
1. Step Function:
f(x)={1if x≥00if x<0f(x) = \begin{cases} 1 & \text{if } x ≥ 0 \\ 0 & \text{if } x < 0 \end{cases}f(x)={10if x≥0if x<0
Used in perceptrons.
2. Sigmoid Function:
f(x)=11+e−xf(x) = \frac{1}{1 + e^{-x}}f(x)=1+e−x1
Smooth curve; output between 0 and 1. Used in binary classification.
3. Tanh Function:
f(x)=tanh(x)f(x) = \tanh(x)f(x)=tanh(x)
Range = (-1, 1); centered around zero.
4. ReLU (Rectified Linear Unit):
f(x)=max(0,x)f(x) = \max(0, x)f(x)=max(0,x)
Efficient for deep networks; reduces vanishing gradient problem.
5. Leaky ReLU:
Allows small negative slope for negative inputs to avoid dead neurons.
SECTION C – Very Long Answer Type Questions (10 Marks each)
(a) Demonstrate K-Nearest Neighbors (KNN) Algorithm for Classification with Example.
Definition:
KNN is an instance-based learning algorithm that classifies a new sample based on the majority label of its K nearest neighbors.
Algorithm Steps:
Choose the number of neighbors KKK.
Compute distance between test sample and all training samples (e.g., Euclidean distance).
Select K nearest samples.
Assign the most frequent class label among them to the test point.
Example:
If K=3 and among the nearest points two belong to class “A” and one to “B,” the new point is classified as
Class A.
Applications:
Image recognition.
Recommendation systems.
(b) Explain Different Layers Used in Convolutional Neural Network (CNN) with Suitable Examples.
1. Input Layer:
Accepts image data (e.g., 28×28×3 for RGB).
2. Convolutional Layer:
Applies filters to extract features like edges or textures.
Y=X∗W+bY = X * W + bY=X∗W+b
Example: Edge detection using 3×3 filter.
3. Activation Layer (ReLU):
Applies non-linearity to feature maps.
4. Pooling Layer:
Reduces spatial dimensions using Max or Average Pooling, retaining important information.
5. Fully Connected Layer:
Combines all extracted features for classification.
6. Output Layer:
Applies softmax activation for multi-class classification (e.g., handwritten digit recognition).