Computer Science · 11th Grade · Artificial Intelligence and Ethics · Weeks 19-27

Machine Learning Fundamentals

Introduction to how computers learn from data through supervised and unsupervised learning.

• Common Core State Standards
• CSTA: 3B-AP-09
• CSTA: 3B-DA-07

About This Topic

Machine learning is a branch of artificial intelligence where systems improve their performance through experience rather than through explicitly programmed rules. Instead of a developer writing every decision a program should make, a machine learning model is trained on examples -- and the patterns it finds become the program. For US 11th graders, this topic bridges math concepts they already know (functions, data, probability) with the practical systems behind recommendation engines, spam filters, and medical diagnosis tools.

The three main learning paradigms map to distinct real-world problem types. Supervised learning powers classifiers like email spam detection, trained on labeled examples. Unsupervised learning groups unlabeled data, as in customer segmentation. Reinforcement learning trains agents through reward signals, used in robotics and game-playing AI. Helping students match the right paradigm to the right problem is a core skill in this unit.

Active learning is especially effective here because ML concepts can feel abstract without hands-on exploration. Running small experiments in tools like Teachable Machine, building intuition by labeling data, and debating which algorithm fits a scenario all accelerate conceptual understanding in ways that lecture alone cannot.

Key Questions

1. Explain the fundamental concept of machine learning and how machines 'learn' from data.
2. Differentiate between supervised, unsupervised, and reinforcement learning paradigms.
3. Analyze real-world problems that can be solved using machine learning.

Learning Objectives

• Compare and contrast the core mechanisms of supervised, unsupervised, and reinforcement learning paradigms.
• Analyze real-world scenarios to identify appropriate machine learning approaches and justify the choice.
• Design a simple supervised learning model using a provided dataset and evaluate its performance.
• Explain the role of data in training machine learning models, distinguishing between labeled and unlabeled data.
• Classify common applications of machine learning, such as spam detection or customer segmentation, into their respective learning paradigms.

Before You Start

Data Representation and Analysis

Why: Students need to understand how data is organized and basic statistical concepts to comprehend how models learn from it.

Functions and Relationships

Why: Machine learning models often represent complex relationships between inputs and outputs, similar to mathematical functions.

Key Vocabulary

Machine Learning	A field of artificial intelligence where computer systems learn from data to improve performance on a task without being explicitly programmed for every step.
Supervised Learning	A type of machine learning where models are trained on labeled datasets, meaning each data point has a known correct output or category.
Unsupervised Learning	A type of machine learning where models are trained on unlabeled datasets, seeking to find patterns, structures, or relationships within the data on their own.
Reinforcement Learning	A type of machine learning where an agent learns to make a sequence of decisions by trying to maximize a reward signal through trial and error in an environment.
Training Data	The dataset used to teach a machine learning model; the model learns patterns and relationships from this data.

Watch Out for These Misconceptions

Common Misconception

Machine learning models understand the world the way humans do.

What to Teach Instead

ML models find statistical patterns in training data. They do not understand context, causality, or meaning. A model that classifies cats with 95% accuracy has no concept of what a cat is -- it has learned a function mapping pixel patterns to labels. Active experiments with adversarial examples that fool classifiers make this concrete in a way that definitions alone cannot.

Common Misconception

More training data always makes a model better.

What to Teach Instead

Unrepresentative, mislabeled, or biased data will degrade model performance regardless of volume. Quality and diversity of training data matter as much as quantity. Students who run their own training experiments in Teachable Machine quickly see this when their model performs poorly on edge cases that weren't well-represented in training.

Common Misconception

Machine learning and artificial intelligence are the same thing.

What to Teach Instead

AI is the broader field of creating systems that perform tasks requiring intelligence. Machine learning is one approach within that field. There are AI systems that are not ML-based -- expert systems and rule-based chatbots -- and not all ML is used for tasks most people intuitively associate with intelligence.

Active Learning Ideas

Collaborative Problem-Solving: Teachable Machine Classifier

Students use Google's Teachable Machine to train an image classifier on three categories of their choosing. They then test the model's accuracy by varying the quantity and diversity of training examples. Pairs document what changed when they added more varied training data and present their findings.

45 min·Pairs

Think-Pair-Share: Sorting by Learning Paradigm

Present five real-world scenarios (detecting fraud, recommending music, training a chess engine, grouping news articles, predicting house prices). Students individually sort them into supervised, unsupervised, or reinforcement learning, then compare with a partner and resolve disagreements by explaining their reasoning before reporting out.

20 min·Pairs

Fishbowl Discussion: Data Bias in ML

A small group in the center discusses a case study of a biased ML model -- such as facial recognition misidentifying darker skin tones -- while the outer ring listens. The outer ring then rotates in to add analysis, and all groups must connect the observed bias back to specific training data choices.

35 min·Small Groups

Gallery Walk: Orange ML Confusion Matrix

Groups use the Orange visual ML tool to run classification on a provided dataset and produce a confusion matrix. Each group posts their results and methodology on the wall; other groups rotate and leave one observation and one question on sticky notes before groups reconvene to respond.

50 min·Small Groups

Real-World Connections

• Netflix uses supervised learning algorithms to predict which movies or shows a user will like based on their viewing history and ratings, personalizing recommendations.
• Banks employ unsupervised learning techniques to identify fraudulent transactions by detecting unusual patterns in customer spending habits that deviate from normal behavior.
• Self-driving car systems utilize reinforcement learning to make driving decisions, such as when to accelerate, brake, or change lanes, based on simulated driving experiences and reward signals for safe navigation.

Assessment Ideas

Quick Check

Present students with three scenarios: 1) Identifying cats vs. dogs in images, 2) Grouping customers into segments based on purchasing behavior, 3) Training a robot to navigate a maze. Ask students to write which learning paradigm (supervised, unsupervised, reinforcement) best fits each scenario and why.

Discussion Prompt

Facilitate a class discussion using the prompt: 'Imagine you are building a system to predict house prices. What kind of data would you need? Would this be supervised or unsupervised learning? Explain your reasoning.'

Exit Ticket

On an index card, have students define one key vocabulary term in their own words and provide one example of its application that was not discussed in class.

Frequently Asked Questions

What is the difference between supervised and unsupervised machine learning?

Supervised learning trains models on labeled examples where the correct answer is known -- such as photos tagged as 'cat' or 'not cat.' Unsupervised learning finds patterns in unlabeled data, grouping similar items without predefined categories. Supervised learning is used when you have a known target outcome; unsupervised learning is used when you're exploring unknown structure in a dataset.

How does a machine learning model actually learn?

During training, the model makes predictions and compares them to correct answers. It then adjusts its internal parameters to reduce the error, a process called optimization. Repeating this across thousands or millions of examples gradually improves accuracy. The specific math typically uses calculus-based gradient descent to find parameter values that minimize prediction error.

What is training data and why does it matter so much?

Training data is the labeled dataset used to teach a model. Its quality, size, and representativeness directly determine what the model learns. If training data contains biases or errors, the model will learn and reproduce them. Data curation is one of the most important and underappreciated parts of building reliable ML systems in practice.

What active learning activities work best for teaching machine learning fundamentals?

Hands-on tools like Google's Teachable Machine let students train their own classifiers without writing code, building intuition about training data quality and model behavior. Sorting real-world scenarios by learning paradigm and debating choices with peers deepens conceptual understanding faster than reading definitions. Both approaches connect abstract vocabulary to concrete outcomes students have directly observed.