Unit-2: Advanced Concepts of Modeling in AI

 Unit-2:  Advanced Concepts of Modeling in AI

AI, ML & Deep Learning – Difference



Artificial Intelligence (AI)

  • AI is the broadest term.

  • It refers to creating machines that can think, learn, and make decisions like humans.

  • Includes: speech assistants, robots, recommendation systems, chatbots.

Machine Learning (ML)

  • ML is a subset of AI.

  • Machines learn from data and improve based on experience.

  • The more data → the smarter the model becomes.

Deep Learning (DL)

  • DL is a subset of ML.

  • Uses Artificial Neural Networks (like human brain neurons).

  • Handles big data, images, videos, voice, etc.

Hierarchy:
👉 AI → ML → DL

2. How ML Works

ML has two main stages:

1. Training Phase

  • The algorithm is fed training data (examples).

  • It learns patterns, rules, relationships from this data.

2. Testing Phase

  • New unseen data (testing data) is used to check how well the model learned.

  • Good models give accurate predictions.

🧩 Important Data Terms

  • Data: Information used for training (images, numbers, sound, text).

  • Features: Properties of data
    e.g., For house price → size, location, bedrooms.

  • Labels: Correct answer
    e.g., “Mango”, “Apple” (for fruit images)

  • Labeled Data: Data + label

  • Unlabeled Data: Only data (no label)

  • Training Data: Data used for learning

  • Testing Data: Data used for accuracy check

3. Types of AI Models

There are two major types:



A. Rule-Based Models



  • These models follow fixed rules written by humans.

  • They do not learn from data.

  • If the situation changes, the rules must be updated manually.

Examples

  • If temperature > 30°C → Turn on fan

  • Chatbots answering fixed questions (FAQs)

Limitations

  • Cannot learn or adapt

  • Fail when new or unexpected situations come

  • Require manual updates frequently

B. Learning-Based Models



  • These models learn patterns automatically from data.

  • Can improve their performance over time.

Examples

  • Spam email detection

  • Product recommendation on Amazon

  • Voice assistants learning speech patterns

4. Types of Learning-Based Models

Machine Learning has 3 main types:



1. Supervised Learning

  • Uses labeled data

  • Learns from examples with correct answers

  • Works like a student learning from a teacher

Used for:

  • Classification (categories)

  • Regression (continuous values)

Examples:

  • Classifying emails → Spam / Not spam

  • Predicting marks from study hours

  • Identifying fruit type from image

  • Handwriting recognition

2. Unsupervised Learning

  • Uses unlabeled data

  • The computer finds patterns on its own

  • No teacher involved

Used for:

  • Clustering

  • Association

Examples:

  • Grouping customers by shopping habits

  • Song recommendation systems

  • Grouping students by performance patterns

  • Identifying dog breeds based on size & colour

3. Reinforcement Learning (RL)

  • Learning by trial and error

  • Gets reward for correct action, penalty for wrong action

  • Ideal for decision-making tasks

Examples:

  • Self-driving cars

  • Robots learning to walk

  • Game-playing AI (chess, Pac-Man)

  • Parking assistance systems




Supervised Learning Models 



A. Classification

  • Output is categorical (fixed groups)

  • Answers are in labels like:

    • Yes / No

    • Hot / Cold

    • Cat / Dog

    • Disease / No disease

Examples

  • Predict if a customer will buy a product

  • Identify handwritten numbers

  • Predict weather as sunny/rainy

B. Regression

  • Output is continuous value

  • Predict numerical data

Examples

  • Predict house price

  • Predict temperature

  • Predict marks in exam

  • Predict sales of a product

6. Unsupervised Learning Models



A. Clustering

  • Groups data automatically

  • No labels

  • Finds hidden patterns

Examples

  • Grouping animals based on features

  • Grouping customers:

    • Frequent buyers

    • Occasional buyers

    • Discount seekers

  • Grouping Netflix users by interest

B. Association

  • Finds relationships between items

  • Used in recommendation systems

Examples

  • “People who buy bread also buy butter

  • Amazon/Flipkart product suggestions

  • Supermarkets analysing purchase patterns

7. Reinforcement Learning 

Features:

  • No labeled data

  • Continuous learning

  • Agent interacts with environment

  • Improves based on rewards/punishments

Examples:

  • Google’s AlphaGo

  • Robots learning tasks

  • Traffic light control systems

  • Autonomous drones

8. Neural Networks 

What is an Artificial Neural Network (ANN)?

ANN is a type of AI model inspired by the human brain and nervous system.
Just like the brain has neurons, ANN has artificial neurons (nodes).
ANN can automatically extract features from data without programmer instructions.
Useful for solving problems where the dataset is very large, such as images.

Key Features of ANN
1. Works Like the Brain
ANN copies how biological neurons send signals.
Nodes in the network work together to make decisions.
2. Automatic Feature Extraction
ANN learns patterns from data by itself.
Programmer does NOT need to tell it which features to use.
3. Each Node Is Like a Small ML Algorithm
Every node processes data, performs its own calculation, and passes the result to the next layer.
4. Good for Large Datasets
Especially useful in tasks such as:
Image recognition
Facial recognition
Chatbots
Price prediction

Structure of an Artificial Neural Network
An ANN has three main layers:

1. Input Layer
First layer.
Job: accept the input data and feed it into the network.
No processing happens here.

2. Hidden Layers
These are middle layers where processing happens.
Called hidden because we don’t directly see what happens inside.
Each node:
Applies weights (importance)
Adds bias
Uses an activation function to transform the data
More hidden layers = more powerful model.

3. Output Layer
Gives the final result to the user.
Like input layer, it does NOT process—just displays the output.

How an ANN Learns 
The ANN learns using a process similar to trial and error:
Data passes from input → hidden → output layer.
The network checks how wrong the output is.
It adjusts weights to reduce the error.
It repeats this many times until it learns correctly.

Real-World Applications
Facial recognition
Customer support chatbots
Vegetable/house price prediction
Self-driving cars
Spam detection 



Comments

Popular posts from this blog

XII UNIT 3 HOW CAN MACHINES SEE?

XII UNIT 2 Data Science Methodology: An Analytic Approach to Capstone Project

UNIT -3 MCQ PART 1 HOW CAN MACHINES SEE?