> For teachers > Elementary School

Lesson plan · 5 days

Becoming a Neural Network

A week-long unit that starts away from the computer, with students acting out a network themselves, and ends with each student designing a network of their own. Every lesson on the elementary page is used, along with two of its example networks.

Grade levels
Grades 3–5
Duration
Five class periods of 45 minutes, across one or two weeks
Class activities
5 activities, one a day, using 8 iiNNoo lessons

Learning objectives

Students will be able to:

  1. 01

    Inputs become outputs

    Describe a neural network as a machine that turns an input number into an output number, one step at a time.

  2. 02

    A weight is a multiplier

    Explain that each connection multiplies the number passing along it, and that a bigger weight makes a bigger output.

  3. 03

    Predict, then check

    Work out what the network will say before it says it, then step through the calculation to check.

  4. 04

    Tune the network

    Change a weight until the network's predictions match what really happened.

  5. 05

    Design a network

    Choose inputs, an output and weights for a prediction from their own life, and explain what each weight means.

Prerequisite knowledge

Before starting, students should be comfortable with:

  • Multiplying a two-digit number by a one-digit number, such as 6 × 25.
  • Adding two numbers to make a total.
  • For Days 2 and 4: knowing that 0.5 is a half and 0.25 a quarter, and reading a decimal such as 1.27. With third graders, do these parts together on the board.
  • Clicking, and typing a number, on a computer or tablet. Students can share one device between two.

A human neural network · 45 minutes

Class Activity: Day 1

Introduction for students

Ask the class for times a computer has made a prediction for them: the weather app, a video site picking the next video, a phone guessing the next word. Explain that these are made by AI, and that this week the class will find out what is inside one, then become one.

Make a prediction together. "A zoo gets about 200 visitors for every hour of sunshine. How many visitors on a day with 3 hours of sun?" Write it on the board as 3 × 200 = 600.

Key Vocabulary

AI
Short for artificial intelligence: computer programs that do things that usually take human thinking, such as making predictions.
Prediction
A guess about something we don't know yet, made from something we do know.
Neural network
A kind of AI that makes predictions by passing numbers along connections and multiplying them.
Input value
The number that goes into the network, such as the hours of sun.
Connection
The line that carries a number from an input to an output. Each connection has its own weight.
Weight
The number on a connection. The input is multiplied by it.
Output value
The number that comes out of the network: its prediction.

Guiding the activity

Set up the human network

Act out a network, with no computer. Three students stand in a row: the Input holds a card with a number, the Connection holds a card saying "× 200", and the Output holds a blank card. The class calls out an hour of sunshine; the Input writes it down and passes it to the Connection, who multiplies and passes the answer to the Output, who announces the prediction.

Run it and change the weight

Run it four or five times with different inputs, then swap the weight card for "× 150" and ask the class what will change. Swap in new students so that most of the class has a turn.

Match it to the app

Then open The Zoo Visitors Predictor on the board and step through it. Ask students to point out which part of the app each of the three volunteers was: the input value, the weight on the connection, the output value. Pairs then step through all four inputs on their own devices.

Post-activity discussion

  1. Which job in our human network was the most important?

    Listen for The connection: it holds the weight, and the weight is what the network knows.

  2. What happened when we changed the weight card?

    Listen for Every prediction changed, though the input and the output did the same job.

  3. How is the app like our human network? How is it different?

    Listen for Same three parts and the same multiplication. The computer does it far faster, and real networks have millions of connections.

Predict, then check · 45 minutes

Class Activity: Day 2

Introduction for students

Recall Day 1 in one sentence from a student: the input is multiplied by the weight to make the output. Today, students will predict what the network will say before it says it.

Share a surprising fact: crickets chirp faster when it is warmer, about 6 more chirps a minute for every degree. Some people use them as a thermometer. Ask how a network could turn the temperature into chirps: with a weight of 6.

Key Vocabulary

Predict
To work out an answer before you see it.
Check
To compare your prediction with the real answer.
Growth rate
How much bigger something gets each year. In the tree network, the weight is the growth rate.

Guiding the activity

Predict the cricket chirps

Pairs open The Cricket Chirp Predictor. Before each output, the app asks for a prediction: students work it out on paper, type it in, then step on to check. Partners take turns: one predicts, the other checks.

Spot the pattern

Listen for students who start adding 30 each time (90, 120, 150, 180) and ask them to explain why that works: each input is 5 degrees warmer, and 5 × 6 = 30.

Explore the tree

Pairs then explore The Tree Growth Predictor, an example with no task to finish. Its weight is 1.27, the tree's growth rate: a 10-meter tree will be 12.7 meters tall next year. Ask students to predict an output before they step to it, and to explain what a weight bigger than 1 does to the input.

Post-activity discussion

  1. How many chirps would the network predict at 10 degrees? At 40?

    Listen for 60 and 240. The network can predict inputs it was never shown.

  2. The tree network's weight is 1.27. What would a weight of 1 mean?

    Listen for The tree stays the same height: the output equals the input.

  3. Why is it useful to predict before you check?

    Listen for It shows whether you understand the network, not just whether you can read its answer.

Fixing a network that's wrong · 45 minutes

Class Activity: Day 3

Introduction for students

So far, every network has had the right weight. Today the weights are wrong, and the class's job is to fix them.

Introduce the target output: the real answer, what actually happened. The network's output is its guess. When the two match, the network is right.

Key Vocabulary

Target output
The real answer, what actually happened. The network's output should match it.
Tuning
Changing a weight, a little at a time, until the outputs match the targets.

Guiding the activity

Model the first fix

Open The Phone Battery Predictor on the board. Step through the first input: the network says a 1-hour video uses 5% of the battery, but the target is 12%. Ask: is 5 too big or too small? Should the weight go up or down? Then hand it over to pairs.

Find the weight

Students click the weight and type a new value. When an output matches its target, the double lines between them turn green. The answer is 12, and it has to work for all four inputs.

Two outputs, two weights

Pairs move on to The Smoothie Stand Predictor, where one input feeds two outputs, each with its own weight. The answers are 4 for smoothies and 2 for popsicles.

Count the tries

Ask students to keep a tally of how many tries each weight took. Some will find a shortcut, dividing the target by the input, and can share it with the class.

Post-activity discussion

  1. How did you know which way to change the weight?

    Listen for If the output was too small, the weight went up; too big, it went down.

  2. Who found the weight in the fewest tries? How?

    Listen for Bring out the two strategies: guess, check and adjust, and dividing the target by the input.

  3. A real AI does what you did, millions of times. What is it doing each time?

    Listen for Measuring how far off its output is, and nudging each weight to make that mistake smaller. This is how AI learns.

Two inputs at once · 45 minutes

Class Activity: Day 4

Introduction for students

Ask: "What decides how many butterflies visit a garden?" Real predictions usually depend on more than one thing, so a network can have more than one input, each with its own connection and its own weight.

Run the human network again with two Inputs, each with its own Connection, and one Output who adds what they are handed. With weights of 1 and 2, a garden with 3 flower beds and 4 bushes gives 3 × 1 + 4 × 2 = 11 butterflies.

Key Vocabulary

Input values
The numbers that go into the network. A network can have more than one, such as flower beds and bushes.
Connection
The line that carries a number from an input to an output. Each connection has its own weight.
Total
With two inputs, the output is each input times its weight, added together.

Guiding the activity

One weight to change

Pairs open The Butterfly Count Predictor. The task is to make each bush count twice as much as each flower bed. The flower bed weight is already 1, so the bush weight needs to be 2. Students change it and confirm it works for all four gardens.

Both weights to find

Next comes The Tomato Harvest Predictor, which asks for both weights, and both are between 0 and 1. The answers are 0.5 for each hour of sun and 0.25 for each liter of water. Encourage students to change one weight at a time.

Put the weights into words

Ask pairs who finish to write down, in words, what each tomato weight means: half a kilogram of tomatoes for each hour of sun, a quarter for each liter of water.

Post-activity discussion

  1. In the tomato garden, which matters more: an hour of sun or a liter of water?

    Listen for An hour of sun: its weight, 0.5, is twice the water's, 0.25.

  2. What does a bigger weight tell us about an input?

    Listen for That the input matters more to the prediction.

  3. Why was this harder than one input?

    Listen for Changing one weight changes every output, so the two weights have to be found together.

Design your own network · 45 minutes

Class Activity: Day 5

Introduction for students

Open The Honey Harvest Predictor on the board, an example with no task to finish. Its two inputs are the flowers growing nearby and the days in the month. Ask the class what each weight means: 0.28 kilograms of honey for each flower, and 0.45 for each day.

Today, each student designs a network of their own, the way the people who made these lessons did: pick something to predict, decide what it depends on, and choose the weights.

Key Vocabulary

Data
Real numbers collected from the world, used to check whether a network's predictions are right.
Design
To plan something before you build it: here, choosing a network's inputs, output and weights.

Guiding the activity

Draw the network

On paper, students draw a network with one or two inputs and one output, from their own life: minutes of reading and pages read, hours of practice and goals scored, scoops of food and days a bag lasts. They label each input, the output and each weight.

Make a table

Next to the drawing, they make a small table of three rows: the inputs, and the output their network predicts. Check that every output is worked out by multiplying each input by its weight and adding.

Swap and predict

Partners swap drawings and predict each other's outputs before looking at the table. Where they disagree, they step through the calculation together.

Build it in the app

With a teacher's help, a student's network can be built in the app: Create, under the logo on the lessons popup, sets its shape and labels, and their table can be uploaded as training data once it is saved from a spreadsheet as a CSV file.

Post-activity discussion

  1. How did you choose your weights?

    Listen for From what they know about the world, the way the cricket weight came from what we know about crickets.

  2. How could you find out whether your weight is right?

    Listen for Collect real data, compare the network's outputs with what really happens, and adjust: Day 3's job.

  3. What would you want a computer to predict, if you could build any network?

    Listen for Let this run. It is a good place to talk about what AI should and shouldn't be used for.

Wrap up

Display the students' network designs around the room as a gallery, and give the class a few minutes to walk around and predict one output on each.

Close the week by asking each student to finish the sentence "A neural network is…" on a sticky note. Strong answers mention inputs, weights and outputs, and that a network learns by checking its answers and adjusting its weights. Students can keep exploring at iinnoo.app, where every lesson is free and needs no account.

Ready to bring iiNNoo to your classroom?

Questions or ideas? Email us at support@iinnoo.app.

Start Learning How AI Learns