AI_for_school_students

🧠 Neural Network Explorer

An interactive, animated Streamlit app that teaches school students how neural networks learn — step by step. Designed for classroom use in rural India, with no prior machine learning knowledge required.

📸 What It Shows

The app walks through 7 guided steps, animating a live neural network:

Step	Concept
🖼️ Step 1	Input — an image is broken into numbers
➡️ Step 2	Forward Pass — signals flow through the network
⚡ Step 3	Activation — neurons decide whether to “fire”
🏁 Step 4	Output — the network makes a prediction
❌ Step 5	Loss — we measure how wrong it was
🔙 Step 6	Backpropagation — the error flows backwards
🔧 Step 7	Weight Update — the network adjusts and learns

A loss curve builds up as students repeat the training steps, letting them literally watch the network get smarter.

🚀 Getting Started

1. Clone or download the files

Make sure you have these two files in the same folder:

neural_network_animation.py
requirements.txt

2. Install Python

You need Python 3.9 or higher. Check your version:

python --version

Download Python from python.org if needed.

3. Install dependencies

pip install -r requirements.txt

4. Run the app

streamlit run neural_network_animation.py

Your browser will open automatically at http://localhost:8501.

🖥️ System Requirements

	Minimum
Python	3.9+
RAM	512 MB
Internet	Only needed for Google Fonts (optional)
OS	Windows, macOS, or Linux

Works on low-end laptops. No GPU required.

🎓 How to Use in the Classroom

Projector Demo (Teacher-led)

Run the app on the teacher’s laptop, project it on screen
Enable Auto-play from the bottom of the page
Set speed to 2–3 seconds per step so students can follow
After all 7 steps, click Next again to repeat — the loss curve will grow

Student Exploration

Each student (or pair) runs the app on their own device
Ask them to try all three input images: Cat 🐱, Dog 🐶, Car 🚗
Have them click through all 7 steps and watch how predictions change
Challenge: “What happens to the loss when you click through many times?”

Discussion Questions

Why does the loss go down over time?
What happens if you set the Learning Rate very high (close to 1.0)?
What does it mean when a connection is red vs blue?
Can you make the network predict “Cat” more confidently?

🎛️ Controls

Control	What it does
◀ Prev / Next ▶	Step through the 7 phases manually
Choose an input image	Switch between Cat, Dog, Car inputs
Learning Rate slider	Controls how big each weight update is
Auto-play toggle	Automatically advances through steps
Speed slider	Sets seconds between auto-play steps
🔄 Reset Network	Resets all weights and the loss curve

🗂️ File Structure

📁 your-folder/
├── neural_network_animation.py   # Main app
├── requirements.txt              # Python dependencies
└── README.md                     # This file

🔬 What’s Under the Hood

The network has this architecture:

Input (3)  →  Hidden 1 (4)  →  Hidden 2 (4)  →  Output (2)

Activation function: ReLU (Rectified Linear Unit)
Output: Softmax probabilities
Loss function: Mean Squared Error (MSE)
Optimiser: Gradient Descent

All weights are initialised randomly and update in real time as students click through the backpropagation step.

🛠️ Troubleshooting

streamlit: command not found

python -m streamlit run neural_network_animation.py

Port already in use

streamlit run neural_network_animation.py --server.port 8502

Fonts not loading (no internet) The app still works fully — it falls back to system fonts automatically.

Slow on old hardware Reduce the number of auto-play steps by using Manual mode (◀ ▶ buttons).

📦 Dependencies

Package	Purpose
`streamlit`	Web UI framework
`matplotlib`	Network diagrams and loss curve
`numpy`	Matrix math for forward pass & backprop
`Pillow`	Image handling (used by Streamlit internally)

📄 Licence

Free to use for educational purposes. Share freely with students and teachers. ❤️

“Every expert was once a beginner.”

This site is open source. Improve this page.