Unit 2: Advanced Concepts of Modeling in AI
Master Rule-Based vs Learning-Based Modeling, Supervised & Unsupervised Learning algorithms, Reinforcement Learning, Artificial Neural Networks (ANN), and step-by-step Perceptron mathematical decisions.
Revisiting AI, ML, and Deep Learning (DL)
Before developing models, it is essential to understand the relationship between Artificial Intelligence, Machine Learning, and Deep Learning:
Artificial Intelligence (AI)
Any technique that enables computers to mimic human intelligence, cognitive problem-solving, and decision-making.
Machine Learning (ML)
Algorithms that enable machines to learn from data and improve at tasks through experience and feedback without being explicitly programmed.
Deep Learning (DL)
Specialized learning approach using Artificial Neural Networks (ANN) with multiple hidden layers that automatically extract features from huge datasets.
Common Data Terminologies:
- Features: Measurable properties or columns of a dataset (e.g., color, size, weight, fuel type, number of bedrooms).
- Labels: The meaningful target tag attached to data that the model aims to predict (e.g., fruit name, spam/not-spam, house price).
- Training Data: Labeled historical examples given to the model during the learning phase (analogous to solved examples shown by a teacher).
- Testing Data: Unseen data used to evaluate model accuracy without providing labels in advance (analogous to an examination test).
- Overfitting: When an AI model memorizes the training data completely, achieving near-100% training score but failing on new, unseen test data.
Types of AI Models: Rule-Based vs Learning-Based
Rule-Based Approach
The developer explicitly codes all rules, conditions, and logical branches: Rules + Data → Answers.
- Static Learning: The machine does not adapt or learn from errors.
- Failure Mode: If input does not match defined rules, it fails or falls back to generic responses.
- Example: E-commerce customer service script-bot with fixed decision trees for order tracking.
Learning-Based Approach
The machine is fed raw data and outputs, and discovers patterns to design its own rules: Data + Answers → Rules.
- Dynamic Adaptation: Adapts to new trends and corrects mistakes through feedback loops.
- Pattern Discovery: Uncovers non-obvious clusters (e.g. grouping 1000 stray dog images by coat texture/size).
- Example: Smart spam filters learning from sender behavior, text keywords, and attachment signatures.
The Three Families of Machine Learning
1. Supervised Learning
Trained on labeled datasets with a known teacher/supervisor.
Classification: Predicts discrete classes (Hot/Cold, Loan Default/Non-Default, Cat/Dog).
Regression: Predicts continuous values (House Price, Temperature, Car Resale Value).
2. Unsupervised Learning
Processes unlabeled datasets to discover intrinsic patterns without guidance.
Clustering: Divides data points into groups by similarity (Supermarket customer segmentation, Spotify songs by tempo/energy).
Association Rule: Discovers co-occurrence relations (Market basket analysis: Bread → Butter).
3. Reinforcement Learning
Agent learns through trial-and-error in dynamic environments to maximize a reward metric.
Mechanism: Positive reward for success, penalty for errors.
Applications: Self-parking cars, robotic humanoid walking, game-playing AI (AlphaGo).
Artificial Neural Networks (ANN) & Perceptron Calculations
Neural networks organize machine learning algorithms into interconnected layers inspired by biological brain neurons. They extract high-level features automatically without manual feature selection.
1. Input Layer
Acquires data and feeds it to the network. No processing occurs here.
2. Hidden Layer(s)
Computes weighted sums and applies non-linear activation functions. Hidden from the user.
3. Output Layer
Delivers the final processed decision or class probability to the user interface.
Case Scenario: “Should I go to the park today?”
Input Variables & Assigned Weights:
• $X_1$: Do I have a jacket? (Weight $W_1 = 1.5$)
• $X_2$: Do I have an umbrella? (Weight $W_2 = 1.0$)
• $X_3$: Is it sunny now? (Weight $W_3 = 3.0$ - Most important)
• $X_4$: Weather forecast for later? (Weight $W_4 = 2.5$)
• Bias $B = 1$ with Caution Weight $W_B = 4.0$
Threshold = 0 (Output > 0 → YES; ≤ 0 → NO)
Mathematical Computations:
Scenario 1: [Jacket=1, Umbrella=0, Sunny=1, Rain=0]
$Output = (1 imes 1.5) + (0 imes 1.0) + (1 imes 3.0) + (0 imes 2.5) - (1 imes 4.0)$
$Output = 1.5 + 0 + 3.0 + 0 - 4.0 = +0.5$
> 0 → Result: Go out to the park! ✔
Scenario 2: [Jacket=0, Umbrella=1, Sunny=0, Rain=1]
$Output = (0 imes 1.5) + (1 imes 1.0) + (0 imes 3.0) + (1 imes 2.5) - (1 imes 4.0)$
$Output = 0 + 1.0 + 0 + 2.5 - 4.0 = -0.5$
< 0 → Result: Do NOT go to the park! ✘
Test Yourself: Official MCQs & Answer Key
Directly from Pages 55–57 of the CBSE Class 10 Facilitator Handbook with official answers from Page 181. Click an option to test your knowledge!
Q1. In which type of machine learning is the data labeled with the desired output?
Q2. An email spam filter that learns to identify spam emails based on labeled examples is an application of:
Q3. A machine learning algorithm that groups similar customer purchases into clusters for recommendation systems uses:
Q4. An AI agent playing a game and learning from its rewards and penalties is an example of:
Q5. Which of the following statements is NOT true about supervised learning?
Q6. In an unsupervised learning scenario, the goal is to:
Q7. Clustering algorithms are commonly used in unsupervised learning for:
Q8. Reinforcement learning is particularly useful for scenarios where:
Q9. Imagine an AI playing a game and learning to win by trial and error. This is an example of:
Q10. Artificial neural networks are inspired by the structure and function of:
Q11. The process of adjusting the weights in a neural network to improve performance is called:
Q12. A neural network with multiple layers of interconnected neurons is called a:
Q13. Neural networks are particularly well-suited for tasks involving:
Q14. Training a neural network often requires:
Reflection Time: Case Studies & Perceptron Conversions
Solved answers for all subjective exercises and scenario questions from Pages 58–60 of the Handbook.
Q1. Give the difference between Rule-based and Learning-based AI models.
Answer: In Rule-Based AI, the developer hardcodes the rules and instructions manually ($Rules + Data o Answers$). The learning is static, and the model cannot adapt if new data violates the rules. In Learning-Based AI, the machine is provided with data and desired outputs ($Data + Answers o Rules$) and dynamically learns the patterns, adjusting its internal algorithms to handle unforeseen variations.
Q2. Differentiate between Classification and Regression models.
Answer: Both are Supervised Learning techniques. Classification predicts a discrete categorical class label (e.g., Yes/No, Spam/Not-Spam, Apple/Orange). Regression predicts a continuous numerical quantity along a range (e.g., House price, temperature, salary, stock value).
Q3. What is Clustering and how is it different from Classification?
Answer: Classification uses predefined classes to assign labeled data points. Clustering is an unsupervised learning technique that groups unlabeled data points based on inherent feature similarities without prior knowledge of class tags (e.g., customer market segmentation).
Identification of Machine Learning Categories in Real-World Scenarios
Q10. Scenario Perceptron Conversions (Page 60)
a) Manager deciding whether to approve a Work-From-Home (WFH) request:
• Inputs: $X_1$ = High remote performance (Weight $W_1 = 3.0$), $X_2$ = No mandatory in-person meetings ($W_2 = 2.0$), $X_3$ = Company policy supports WFH ($W_3 = 2.5$), $X_4$ = Mutually beneficial ($W_4 = 1.5$).
• Bias: Bias weight $W_B = 4.0$ (company preference for in-office collaboration).
• Decision Rule: If $\sum W_i X_i - W_B > 0$, WFH is approved.
b) Homeowner deciding whether to invest in Solar Panels:
• Inputs: $X_1$ = High average sunlight ($W_1 = 3.5$), $X_2$ = Government rebates available ($W_2 = 2.5$), $X_3$ = Increases home value ($W_3 = 1.5$), $X_4$ = Environmental benefits ($W_4 = 2.0$).
• Bias: Bias weight $W_B = 4.5$ (high upfront installation expenditure hurdle).
• Decision Rule: If $\sum W_i X_i - W_B > 0$, invest in solar panels.
Unit 2 Key Takeaways at a Glance
Frequently Asked Questions
Book a Free Demo Class
Master Perceptron calculations, Artificial Neural Networks, and Classification vs Regression with top CBSE AI mentors. Free 45-minute live trial class!