Lesson planPaper Neural Network: Hand-Calculate How a Neuron Decides
STEAM45 minutesGrades: 10th
Overview
Train a paper neuron by hand and watch it learn.
Description
Your teen builds a one-neuron 'picnic predictor' on poster paper, with three inputs (sunny, warm, weekend), a weight on each string and a threshold of 3. They run it on six sample days, apply a simple learning rule every time it is wrong, and watch the weights change from 1, 1, 1 to 1, 1, 2 until it gets every day right.
Objective
Your teen will calculate a neuron's weighted sum, compare it to a threshold, and apply a learning rule to adjust weights until the neuron correctly classifies a set of training examples.
Materials
| Item | Quantity |
|---|
| Large sheet of paper or poster board | 1 |
| Pieces of colored string or yarn, about 30 cm each | 3 |
| Roll of tape | 1 |
| Index cards | 10 |
| Calculator | 1 |
| Pencil and eraser | 1 each |
Vocabulary
Artificial neuron
A simple math unit that multiplies each input by a weight, adds the results and gives an output based on whether the total reaches a threshold.
Weight
A number that sets how much one input matters to the neuron's decision.
Threshold
The total the weighted sum must reach for the neuron to output yes (1).
Perceptron learning rule
A training rule that nudges weights up or down by a fixed step whenever the neuron gives a wrong answer.
Preparation
- Copy the six Training Days from Additional Sections onto 6 of the index cards, one day per card.
- Label 2 more index cards 'Weights' and 'Threshold = 3', and keep the last 2 blank for testing.
Introduction
Ask: when you decide whether to go out, do you weigh some things more than others? Today you will build a 'brain cell' that makes one decision, picnic or no picnic, and teach it by correcting its mistakes.
Which matters more to you for a picnic: sun, warmth, or it being a weekend? Why?
What do you think it means for a computer to 'learn' something?
Steps
1Step 1
Build the neuron
(8 min) On the large sheet of paper your teen draws three input circles labeled Sunny, Warm and Weekend on the left and one big neuron circle on the right, then tapes one of the 3 pieces of string from each input to the neuron. They write a starting weight of 1 on a piece of tape on each string and tape the 'Threshold = 3' index card next to the neuron.
2Step 2
Learn the math
(7 min) Explain the rule: each input is 1 (yes) or 0 (no), multiply each by its weight, add them up, and the neuron says Picnic (1) if the sum is 3 or more. Your teen tests it on Day A from the index cards, writing the sum with the pencil on the 'Weights' card, and checks it against the Day A answer.
3Step 3
Train with the learning rule
(15 min) Your teen runs all 6 training-day index cards in order. Each time the neuron is wrong they apply the learning rule: if it said no but should say yes, add 1 to the weight of every input that was 1; if it said yes but should say no, subtract 1 from those weights. They rewrite the tape labels after every change and log each day's sum, output and new weights on the 'Weights' card, using the calculator to check sums.
4Step 4
Run a second pass
(8 min) Your teen runs the 6 index cards a second time with the new weights and counts the mistakes, which should now be zero. Check their trace against the Answer Key and have them explain in their own words why the Weekend string ended up heavier than the others.
5Step 5
Test new days
(7 min) You write two new days on the 2 blank index cards, such as sunny and weekend but cold, and your teen predicts the output with the trained weights before you reveal what a real person would choose. Finish by asking how a real neural network with millions of weights is the same idea and what makes it different.
Conclusion
Ask your teen: 'Nobody told the neuron that weekends matter most. How did it figure that out?' That is training: wrong answers push the weights until the patterns in the data show up in the numbers.
Make your own lesson plan in minutes
Describe a topic and Resource Portal AI writes the objectives, materials and steps for your group.
Try Resource Portal AI