Class 10 · Unit 2: Modelling · Lesson 2.4 · 40 min

Neural networks, and how AI makes a decision

Same sky, same forecast. Bilal goes to play cricket; Hiba stays home. Which of them is wrong?

Today you will: Build one neuron in a spreadsheet · Change its weights and change its mind · Be a node in a human neural network

Story

Saturday. The sun is out over the ground behind the school, but the forecast says rain by four.

Bilal: “I'm going. It's sunny now. I've got my jacket.”

Hiba: “I'm not. It says rain later, and I haven't got an umbrella.”

Bilal: “Same sky. Same forecast. One of us is wrong.”

Their teacher, passing, stops. "Neither of you is wrong. You're weighing the same things differently."

Watch

One neuron, then many — — scan code 10.2.4 in the printed book.

Warm up your fingers

Skill: a formula that multiplies two columns (5 min)

On typing.com: the symbols lessons.
Type three times, eyes on the screen:

=SUMPRODUCT(B2:B6;C2:C6)
Goal this term: 38–44 WPM at 95%.

On the laptop

You need: LibreOffice Calc · neuron.ods

Mission 1: Build one neuron (8 min, in pairs)

☐1
Open neuron.ods. Column A has the four questions and Bias. Column B is the input: 1 for yes, 0 for no. The bias input is always 1. Column C holds the weights: 3, −2, 1.5, 1, −2.
☐2
Set Saturday's inputs: sunny 1, rain forecast 1, jacket 1, umbrella 0.
☐3
Predict: go or stay? Write it in E1.
☐4
In C8, add it all up: =SUMPRODUCT(B2:B6;C2:C6). In C9, decide: =IF(C8>0;"GO";"STAY"). Were you right?

Mission 2: Same day, different weights (5 min)

☐5
You are now Hiba: more cautious. Change the bias weight from −2 to −4. Same inputs — what does the neuron decide now?
☐6
Find the smallest change to one weight that turns Hiba's STAY back into GO. Write it in E3.

Mission 3: The human neural network (15 min, whole class, away from the laptops)

☐7
Your teacher places 20 of you as a network: 7 in the input layer, 6 in hidden layer 1, 6 in hidden layer 2, and 1 in the output layer. Everyone else is an observer.
☐8
Only the input layer sees a picture. No talking, all game. Each node writes one word per chit and passes it forward, as your teacher explains.
☐9
The output node reads the chits and guesses the picture. Then the picture is revealed.

Finished early?

Online: open playground.tensorflow.org, press play, and watch a real network change its weights as it learns. What happens when you add a hidden layer?

No laptop today?

Do Parts 1 and 2 on the board: the class calls out 1s and 0s, and pairs do the multiplying on paper. Then play the human neural network with the whole period to spare.

Now you know

One artificial neuron: four yes-or-no inputs, each multiplied by a weight, plus a bias, added up and compared with zero to decide.

A neuron weighs its inputs. Each input is multiplied by a weight — how much it matters — then everything is added with a bias and compared with a threshold.
Different weights, different decisions. Bilal and Hiba saw the same sky. Their weights differed.
A neural network is layers of neurons: an input layer, hidden layers that do the working, an output layer that answers.
It learns by adjusting weights. It guesses, measures how wrong it was, and nudges every weight — thousands of times.
Its big advantage: it finds useful features by itself, which is why it suits huge data like images. CBSE names ANN and CNN; CNNs are Unit 4.

Debate it

In the game, no single person knew the picture except the input layer, and nobody could explain how the output node reached its guess.

A real network has millions of weights. Could anyone explain why it made one decision?
If a network like that turned down your scholarship, what would you want to be told?

Check yourself — practice, not a test

1. Put the layers of a neural network in order, from data to answer:
- 1. Input layer
- 2. Output layer
- 3. Hidden layers
2. In a neuron, what does a weight say?
- ○ How heavy the input is
- ○ How much that input matters to the decision
- ○ How many inputs there are
- ○ Whether the answer is right
3. A neural network learns by adjusting its weights after its mistakes.
- ○ True
- ○ False
4. The kind of neural network built for images, which Unit 4 explores, is a ______ neural network.
5. Sunny 1, rain forecast 1, jacket 1, umbrella 0, with weights 3, −2, 1.5, 1 and a bias of −2. What is the total?
- ○ 2.5
- ○ 0.5
- ○ −1.5
- ○ 4.5

Remember

A neuron weighs. A network learns by changing its weights.

Story

Saturday. The sun is out over the ground behind the school, but the forecast says rain by four.

Bilal: “I'm going. It's sunny now. I've got my jacket.”

Hiba: “I'm not. It says rain later, and I haven't got an umbrella.”

Bilal: “Same sky. Same forecast. One of us is wrong.”

Their teacher, passing, stops. "Neither of you is wrong. You're weighing the same things differently."

On the laptop

You need: LibreOffice Calc · neuron.ods

Mission 1: Build one neuron (8 min, in pairs)

☐1
Open neuron.ods. Column A has the four questions and Bias. Column B is the input: 1 for yes, 0 for no. The bias input is always 1. Column C holds the weights: 3, −2, 1.5, 1, −2.
☐2
Set Saturday's inputs: sunny 1, rain forecast 1, jacket 1, umbrella 0.
☐3
Predict: go or stay? Write it in E1.
☐4
In C8, add it all up: =SUMPRODUCT(B2:B6;C2:C6). In C9, decide: =IF(C8>0;"GO";"STAY"). Were you right?

Mission 2: Same day, different weights (5 min)

☐5
You are now Hiba: more cautious. Change the bias weight from −2 to −4. Same inputs — what does the neuron decide now?
☐6
Find the smallest change to one weight that turns Hiba's STAY back into GO. Write it in E3.

Mission 3: The human neural network (15 min, whole class, away from the laptops)

☐7
Your teacher places 20 of you as a network: 7 in the input layer, 6 in hidden layer 1, 6 in hidden layer 2, and 1 in the output layer. Everyone else is an observer.
☐8
Only the input layer sees a picture. No talking, all game. Each node writes one word per chit and passes it forward, as your teacher explains.
☐9
The output node reads the chits and guesses the picture. Then the picture is revealed.

Finished early?

Online: open playground.tensorflow.org, press play, and watch a real network change its weights as it learns. What happens when you add a hidden layer?

No laptop today?

Do Parts 1 and 2 on the board: the class calls out 1s and 0s, and pairs do the multiplying on paper. Then play the human neural network with the whole period to spare.

Now you know

A neuron weighs its inputs. Each input is multiplied by a weight — how much it matters — then everything is added with a bias and compared with a threshold.

Different weights, different decisions. Bilal and Hiba saw the same sky. Their weights differed.

A neural network is layers of neurons: an input layer, hidden layers that do the working, an output layer that answers.

It learns by adjusting weights. It guesses, measures how wrong it was, and nudges every weight — thousands of times.

Its big advantage: it finds useful features by itself, which is why it suits huge data like images. CBSE names ANN and CNN; CNNs are Unit 4.

Debate it

In the game, no single person knew the picture except the input layer, and nobody could explain how the output node reached its guess.

A real network has millions of weights. Could anyone explain why it made one decision?
If a network like that turned down your scholarship, what would you want to be told?

Check yourself — practice, not a test

1. Put the layers of a neural network in order, from data to answer:

1. Input layer
2. Output layer
3. Hidden layers

2. In a neuron, what does a weight say?

○ How heavy the input is
○ How much that input matters to the decision
○ How many inputs there are
○ Whether the answer is right

3. A neural network learns by adjusting its weights after its mistakes.

○ True
○ False

4. The kind of neural network built for images, which Unit 4 explores, is a ______ neural network.

5. Sunny 1, rain forecast 1, jacket 1, umbrella 0, with weights 3, −2, 1.5, 1 and a bias of −2. What is the total?

○ 2.5
○ 0.5
○ −1.5
○ 4.5