Define PR? Explain different stages of pattern recognition? Define dimensionality reduction?

Why do we reduce dimension of

Pattern Recognition (PR) is the process of identifying or feature vectors?
classifying objects, patterns, or signals based on their features and Dimensionality reduction is the process of reducing the number of
regularities. It is a key area in machine learning, computer vision, features (or dimensions) in a dataset while preserving as much of the
and data analysis, enabling systems to recognize and interpret data relevant information as possible.
from the real world. We reduce the dimensions of feature vectors to:
Stages of Pattern Recognition: The pattern recognition process *Simplify data visualization and interpretation.
can be broken down into the following stages: *Remove noise and redundant features.
Sensing (Data Acquisition) Objective: Collect raw data from *Improve computational efficiency.
sensors or data sources. *Prevent overfitting in machine learning models.
3. Feature Extraction - Objective: Identify and extract meaningful
features or attributes from the data that help differentiate between
patterns. Define classification? Explain different classification tasks?
Classification/Decision Making- Objective: Use extracted Classification is a supervised learning task where a model predicts
features to assign the input to one of the predefined categories or the category of input data.
classes. Types of Classification Tasks
Postprocessing - Objective: Refine and interpret the output of the 1. Binary Classification: Two classes (e.g., spam vs. not
classification stage. spam).
Evaluation- Objective: Assess the performance of the pattern 2. Multiclass Classification: More than two classes (e.g., fruit
recognition system. types).
3. Multilabel Classification: Multiple labels per instance (e.g.,
Explain any two real time examples of pattern recognition? tags for a news article).
Facial Recognition in Security Systems 4. Imbalanced Classification: Uneven class distribution (e.g.,
Use Case: Unlocking devices, identifying individuals in fraud detection).
surveillance footage, or granting access to secure areas. 5. Ordinal Classification: Ordered classes (e.g., ratings: poor,
How It Works: good, excellent).
Data Acquisition:: A camera captures an image or video of a face.
Preprocessing::- The system adjusts the image for lighting, scales
it to a standard size, and aligns facial features.
Real-World Examples:
Explain the importance of vector, matrix and tensor at feature
*Apple Face ID: Unlocking iPhones using 3D facial recognition.
*CCTV Surveillance: Identifying persons of interest in public extraction process?
places. Importance of Vectors, Matrices, and Tensors in Feature
Spam Email Detection Extraction
Use Case: Filtering unwanted or malicious emails in email inboxes Vector: Represents individual data points (e.g., pixel intensity,
(e.g., Gmail, Outlook). numerical features) and enables calculations like distance and
How It Works: transformation.
Data Acquisition::- Incoming emails are collected along with Matrix: Organizes multiple data points, used for linear
metadata (sender, subject, body content). transformations, dimensionality reduction (e.g., PCA), and image
Preprocessing::- Text from emails is cleaned by removing stop data representation.
words, punctuation, and irrelevant characters. Tensor: Handles high-dimensional data (e.g., videos, 3D images),
Real-World Examples:Google Gmail: Uses machine learning to essential for deep learning and convolution operations.
filter 99.9% of spam emails.
Outlook's Focused Inbox: Prioritizes legitimate emails over
potential spam How do you define similarity between two feature
vectors? Explain any one technique to analysis the
distance between two features vectors.
Define features? Explain different types of learning mechanism to
extract features automatically?
Defining Similarity
Features are individual measurable properties or characteristics of Similarity between feature vectors measures how
data that are used as inputs for machine learning models. They help close they are, often using distance metrics.
the model identify patterns or make predictions.
Types of Learning Mechanisms to Extract Features
Euclidean Distance
Automatically: Formula: d(A,B)=∑i=1n(ai−bi)2d
Supervised Learning::*Uses labeled data to learn features that Steps:
correlate with output labels.*Example: CNNs in image *Subtract corresponding elements of the vectors.
classification.
Unsupervised Learning::: *Uses unlabeled data to discover
*Square the differences, sum them up, and take the
patterns or structures.*Example: PCA for dimensionality reduction. square root.
Self-Supervised Learning::*The model generates its own Use:
labels)…*Example: BERT for text learning by predicting masked Common in clustering (e.g., K-means) and
words.
Semi-Supervised Learning:*Combines both labeled and nearest-neighbor tasks.
unlabeled data to learn features.*Example: Speech recognition Effective with normalized data.
systems using a mix of labeled and unlabeled audio.
Define image transformation? Explain the steps to Prove the given two datasets of pizza price are
transform a image from V space to U space with matrix M. identical using appropriate feature selection
Image Transformation techniques. Here the two dataset contains pizza
Image transformation changes an image's representation
using mathematical operations (e.g., rotation, scaling). price of different cities in USA (US $ ) and Mexico(
Steps to Transform from V space to U space using Pesos):
Matrix M To prove that the two datasets of pizza prices (in US
*Represent the Image in V space: The image is Dollars and Mexican Pesos) are identical, we will use
represented as a vector in the V space.
*Apply Matrix M: Multiply the vector V by feature selection techniques and verify whether the
transformation matrix M: U=M⋅V two datasets are linearly related by a constant
*Transformation Operations: Matrix M applies conversion factor.
operations like scaling or rotation. Given data:
*Result: The transformed image is represented in the U
space. Approach:
We hypothesize that the two datasets are related
by a constant conversion factor k such that:
Explain discrete random variables and continuous random Y=k⋅X
variables with examples? where k is the conversion rate from US Dollars
Discrete Random Variables to Mexican Pesos.
 Take distinct, separate values. To prove this, we calculate the value of k for each
 Examples:
data point.
o The number of heads in 5 coin tosses
(values: 0, 1, 2, 3, 4, 5).
Calculate the Conversion Factor k = Y \ X
o The number of students in a class US $ (X) Pesos (Y) k=Y\X
(values: 0, 1, 2, …). 1---------------------18.81---------------18.81
Continuous Random Variables 2---------------------37.62----------------18.81
 Can take any value within a range, including 3---------------------56.43----------------18.81
fractions or decimals. Verify the Relationship:
 Examples: For each X, calculate Ycalculated=k⋅X using
o The height of a person (e.g., 5.6 feet,
k=18.81, and compare it with the given Y.
5.61 feet, etc.).
o The time taken to run a race (e.g., 12.5 Conclusion:
seconds, 12.55 seconds). Since the conversion factor k=18.81 is constant
across all data points, and the calculated
Ycalculated values match the given Y values
perfectly, the two datasets are identical.
Define clustering? Explain any one clustering The relationship between the datasets is:
technique? Y = 18.81⋅X
Clustering where 1US Dollar=18.81Mexican Pesos.
Clustering is an unsupervised learning technique
where similar data points are grouped together
into clusters based on their features.
K-Means Clustering Explain probability density function with proper
Initialization: Choose kkk initial centroids randomly. illustration?
Assign Points: Assign each data point to the nearest
centroid. A Probability Density Function (PDF) is a statistical
Update Centroids: Recalculate the centroids as the function that describes the likelihood of a continuous
mean of the points in each cluster. random variable taking a particular value. The PDF is
Repeat: Repeat steps 2 and 3 until the centroids used for continuous probability distributions and
stabilize. gives the relative likelihood of different outcomes
K-means is simple and widely used for grouping within a given range.
data into kkk distinct clusters.