Unit 2: Advanced Concepts of Modelling in AI
Class 10 Artificial Intelligence notes — clear definitions of Supervised, Unsupervised & Reinforcement Learning, Neural Networks and Perceptrons, plus solved MCQs, Assertion-Reasoning and case-study questions with exam-ready answers.
📘 Concept Notes
2.1 Revisiting AI, ML, DL
- Artificial Intelligence (AI): Any technique that enables computers to mimic human intelligence. It works on algorithms and data fed to it to produce a desired output.
- Machine Learning (ML): Enables machines to improve at tasks with experience — it learns from new data, including its own past mistakes, and uses this learning for the next iteration.
- Deep Learning (DL): Enables software to train itself using vast amounts of data, with multiple ML algorithms working together. It is the most advanced of the three.
- Relationship: AI is the umbrella term. ML is a subset of AI. DL is a subset of ML (funnel: AI → ML → DL).
Examples of ML: Object Classification, Anomaly Detection.
Examples of DL: Object Identification, Digit Recognition.
Common Data Terminology
- Data: Information in any form (e.g., a table of fruits).
- Features: The columns of a dataset (e.g., name, color, size).
- Labels: Special features that carry the "answer" — attaching meaning to data based on the problem context.
- Labeled Data: Data with a tag/label attached.
- Unlabeled Data: Raw data with no tag attached.
- Training Dataset: Examples given to the model to learn from.
- Testing Dataset: Used to check the model's accuracy, performed without labels and then verified against them.
2.2 Modelling
Rule-Based Approach
The AI modelling approach where the relationship or patterns in data are defined by the developer. The machine strictly follows the rules and performs its task accordingly.
- Example: A rule-based chatbot answering FAQs using a fixed decision tree of if-then rules.
- Limitation: Learning is static — the model cannot adapt to new or changed data and fails outside its trained rules.
Learning-Based Approach
The AI modelling approach where the machine learns by itself from data and feedback, rather than being explicitly programmed.
- Example: A learning-based spam filter that learns from labeled emails and adapts over time.
- Advantage: It modifies itself as the underlying data changes.
Types of AI Models (Hierarchy)
AI Models
├── Rule Based
└── Learning Based
├── Machine Learning
│ ├── Supervised Learning
│ ├── Unsupervised Learning
│ └── Reinforcement Learning
└── Deep Learning
├── Artificial Neural Networks (ANN)
└── Convolutional Neural Networks (CNN)
Categories of Machine Learning
1. Supervised Learning
A type of machine learning in which the model is trained on a labeled dataset — a dataset where every input is tagged with the correct output. The model learns the relationship between input and output and uses it to predict outcomes for new data.
Example: Predicting the currency of a coin from its weight (Feature = weight, Label = currency).
2. Unsupervised Learning
A type of machine learning in which the model is trained on an unlabeled dataset. The model identifies hidden patterns, similarities, and relationships within the data on its own, without any guidance.
Example: Grouping supermarket customers into clusters based on purchase patterns, with no predefined "customer type" labels.
• Social media tagging a friend in a photo → Supervised (tagged photos act as labels).
• OTT recommendations from watch history → Unsupervised (patterns discovered from likes, no predefined labels).
• Flagging suspicious bank transactions (fraud undefined) → Unsupervised (no "fraud"/"not fraud" labels exist; outliers are flagged).
3. Reinforcement Learning
A type of machine learning in which a model learns to make decisions through trial and error, receiving rewards for correct actions and penalties for incorrect ones, in order to maximize a reward metric — without being explicitly programmed for the task.
Example: Parking a car, humanoid walking, or a machine learning to identify a fruit through repeated positive/negative feedback.
| Learning Type | Goal |
|---|---|
| Supervised | Determine relationships through training |
| Unsupervised | Discover new patterns |
| Reinforcement | Learn by rewarding actions |
Sub-Categories of Supervised Learning
Classification Model
A supervised learning model that works on discrete data and predicts a category/class as output.
Examples: Predicting "hot"/"cold" weather; classifying emails as "spam"/"not spam."
Regression Model
A supervised learning model that works on continuous data and predicts a numeric value as output.
Examples: Predicting temperature; predicting house price (Features: bedrooms, carpet size, garage area → Label: price); used-car price prediction.
• Predicting loan eligibility (yes/no) → Classification (discrete/binary output).
• Predicting weather for the next 24 hours → Regression (continuous range).
Sub-Categories of Unsupervised Learning
Clustering
Groups data points based on the similarity of characteristics, without using any predefined classes or labels.
Difference from Classification: Classification uses predefined classes; Clustering discovers groupings on its own.
Example: Grouping songs by tempo and intensity to predict listener preference.
Association
An unsupervised learning method used to find interesting relationships between variables in a dataset.
Example: Customers who buy bread are likely to also buy butter (market basket analysis).
Sub-Categories of Deep Learning
- Artificial Neural Networks (ANN):Artificial Neural networks are modelled on the human brain and nervous system. They are able to automatically extract features without input from the programmer. Every neural network node is essentially a machine learning algorithm. It is useful when solving problems for which the data set is very large.
- Convolutional Neural Networks (CNN): Convolutional Neural Network is a Deep Learning algorithm which can take in an input image, assign importance (learnable weights and biases) to various aspects/objects in the image and be able to differentiate one from the other
2.3 Artificial Neural Networks
A Neural Network is a system loosely modelled on the structure and functioning of the human brain. It automatically extracts data features without needing input from the programmer, and is organized into multiple layers of interconnected nodes.
Structure — Three Layers
- Input Layer: Acquires data and feeds it into the network. No processing occurs here.
- Hidden Layer(s): All processing/computation occurs here using weights and biases. The result at each node passes through an activation function before moving to the next layer. There can be multiple hidden layers depending on task complexity.
- Output Layer: Delivers the final result to the user. No processing occurs here either — it exists purely for the user interface.
Learning happens through trial and error — weights are adjusted based on the error between the desired output and the actual network output.
Real-world applications: facial recognition, customer-support chatbots, vegetable price prediction.
Perceptron (Decision-Making Unit)
A Perceptron is the basic computational unit of a neural network. It takes multiple inputs (X1, X2, X3…), each multiplied by an assigned weight (W1, W2, W3…), adds a bias term (with its own weight WB), sums all these weighted values, and compares the sum against a threshold to produce a final output/decision.
Weight values come from experience and personal preference — there is no single "right" set of weights, which is why different people can reach different decisions in the same situation.
Activity: Human Neural Network Game
A classroom activity that simulates a neural network using students as nodes across the Input Layer, Hidden Layer 1, Hidden Layer 2, and Output Layer. Information (words on chits) passes silently, layer to layer, until the Output Layer guesses the original image — demonstrating how data flows and gets transformed through a real neural network.
✅ Practice Zone — MCQs, Test Yourself, Assertion-Reasoning & Reflection Q&A
All options are shown upfront. Tap + on any question to reveal the correct answer.
Multiple Choice Questions
1. In which type of machine learning is the data labeled with the desired output?
2. An email spam filter that learns to identify spam emails based on labeled examples is an application of:
3. A machine learning algorithm that groups similar customer purchases into clusters for recommendation systems uses:
4. An AI agent playing a game and learning from its rewards and penalties is an example of:
5. Which of the following statements is NOT true about supervised learning?
6. In an unsupervised learning scenario, the goal is to:
7. Clustering algorithms are commonly used in unsupervised learning for:
8. Reinforcement learning is particularly useful for scenarios where:
9. Imagine an AI playing a game and learning to win by trial and error. This is an example of:
10. Artificial neural networks are inspired by the structure and function of:
11. The process of adjusting the weights in a neural network to improve performance is called:
12. A neural network with multiple layers of interconnected neurons is called a:
13. Neural networks are particularly well-suited for tasks involving:
14. Training a neural network often requires:
Assertion & Reasoning
A&R-1. Assertion (A): Unsupervised Learning is a type of learning without any guidance.
Reasoning (R): Unsupervised learning models work on unlabeled datasets, where the data fed into the machine is random and the person training the model may not have any prior information about it.
A&R-2. Assertion (A): Information processing in a neural network relies on weights and biases assigned to nodes.
Reasoning (R): These weights and biases determine how strongly a node is influenced by its inputs and its overall contribution to the next layer.
Reflection Time — Exam-Style Q&A
Q1. Give the difference between rule-based and learning-based AI models.
- Rule-Based Approach: An AI modelling technique where the relationship or patterns in data are defined by the developer. The machine strictly follows the pre-written rules.
- Learning-Based Approach: An AI modelling technique where the machine learns by itself from data, adapting automatically as data changes.
| Basis | Rule-Based | Learning-Based |
|---|---|---|
| Logic defined by | Developer | Machine (learns on its own) |
| Adaptability | Static — cannot adapt | Dynamic — adapts to new data |
| Example | Rule-based FAQ chatbot | Spam filter learning from labeled emails |
Q2. What is supervised, unsupervised and reinforcement learning? Explain with examples.
Supervised Learning: A type of machine learning where the model is trained on a labeled dataset — every input is tagged with the correct output. Example: Predicting a coin's currency from its weight.
Unsupervised Learning: A type of machine learning where the model is trained on an unlabeled dataset and identifies hidden patterns on its own. Example: Grouping customers by purchase pattern.
Reinforcement Learning: A type of machine learning where the model learns through trial and error, using rewards and penalties to maximize a reward metric. Example: An AI learning to park a car.
Q3. What is clustering and how is it different from classification?
Clustering is an unsupervised learning technique that groups data points based on the similarity of their characteristics, without using predefined classes or labels.
| Basis | Classification | Clustering |
|---|---|---|
| Learning type | Supervised | Unsupervised |
| Classes | Predefined | Discovered on its own |
| Example | Spam / not spam | Grouping songs by tempo |
Q4. Explain neural networks. Also give functions of three layers of neural networks.
A Neural Network is a system loosely modelled on the human brain that automatically extracts data features without needing programmer input, organized into layers of interconnected nodes.
- Input Layer: Acquires data and feeds it into the network. No processing occurs here.
- Hidden Layer(s): All processing occurs here using weights, biases, and an activation function.
- Output Layer: Delivers the final result to the user. No processing occurs here either.
Q5. Differentiate between classification and regression model.
| Basis | Classification Model | Regression Model |
|---|---|---|
| Definition | Predicts a category/class from discrete data | Predicts a numeric value from continuous data |
| Example | Hot/cold weather | House price prediction |
Q6. What is neural network? Give the functioning of its three layers.
Same as Q4 above — a Neural Network automatically extracts features and is organized into Input Layer (data intake, no processing), Hidden Layer (processing via weights/biases), and Output Layer (delivers result, no processing).
Q7. Case Study — Identify the type of learning for each scenario.
a) A company predicts customer churn using labeled data of churned/non-churned customers.
Answer Supervised Learning — Explanation: Supervised learning uses a dataset where every input has a known correct output. Here, customers are already labeled as "churned" or "not churned."
b) A social media platform groups users by interest/behavior with no predefined categories.
Answer Unsupervised Learning — Explanation: No predefined labels exist, so the model discovers natural groupings on its own.
c) An autonomous vehicle learns to navigate via rewards for safe driving and penalties for violations.
Answer Reinforcement Learning — Explanation: Learning is driven by a reward-and-penalty feedback loop, the defining feature of reinforcement learning.
d) Finding patterns in patient data with no predefined treatment labels.
Answer Unsupervised Learning — Explanation: With no predefined labels, the model must independently identify patterns among patients.
e) Detecting anomalies using labeled examples of normal and anomalous machine behavior.
Answer Supervised Learning — Explanation: The dataset explicitly labels behavior as "normal" or "anomalous," so the model learns from these labels.
Q8. Case Study — Identify the type of model for each scenario.
a) Predicting loan "default"/"non-default" status.
Answer Classification Model — Explanation: The output is one of two fixed discrete categories.
b) Predicting house selling price from size, location, bedrooms, bathrooms.
Answer Regression Model — Explanation: "Price" is a continuous numeric value.
c) Segmenting customers into groups based on purchasing behavior.
Answer Clustering Model — Explanation: Customers are grouped by similarity with no predefined categories.
d) Finding which products are commonly bought together.
Answer Association Model — Explanation: This is a classic association-rule / market-basket problem.
Q9. Case Study — Healthcare provider predicting hospital stay length.
a) Predict whether a patient will have a short or long hospital stay.
Answer Classification Model — the output is a discrete category (short/long).
b) Predict the number of days a patient will stay.
Answer Regression Model — the output is a continuous numeric value.
c) Segment patients into groups with similar characteristics.
Answer Clustering Model — patients are grouped by similarity, no predefined labels.
d) Identify patterns in patient treatments and outcomes.
Answer Association Model — the goal is to find relationships between treatments and outcomes.
Q10. Define a Perceptron.
A Perceptron is the basic computational unit of a neural network. It takes multiple inputs, each multiplied by an assigned weight, adds a bias term (with its own weight), sums all these weighted values, and compares the sum against a threshold to produce a final output/decision.