Explain It In Pizza

Learn how AI is trained — one slice at a time 🍕

Click any slice to jump to that section →

🍅 Training Data = Pizza Ingredients

Feed the model examples to learn from, just like gathering ingredients for pizza

Try It Yourself

Drag ingredients into the bowl to build your training dataset. Notice how the counter increases with good ingredients but decreases with bad ones (the rubber duck!). The model learns better with more diverse, quality examples.

🏭 Neural Network = Pizza Assembly Line

Data flows through interconnected layers, transforming at each step

Try It Yourself

Press "Send an Order!" to watch data flow through the neural network. Each layer (node) processes information and passes it forward, transforming raw inputs into meaningful outputs — just like stations on an assembly line.

👨‍⚖️ Loss Function = Pizza Quality Judge

A scoring system that measures how far your pizza is from perfect

Try It Yourself

Use the sliders to adjust your pizza to match the target. The judge scores the difference and reacts based on quality. Lower loss = happier judge! This is how the model knows if it's on the right track.

⛰️ Gradient Descent = Following the Recipe Downhill

Rolling down the loss landscape to find the best recipe

Try It Yourself

Adjust the learning rate and watch the pizza ball roll downhill to the best recipe. Too slow, and you're stuck; too fast, and you overshoot. The goal is to find the sweet spot where loss is minimized.

🧠 Overfitting = One-Track Chef

A model that memorizes training data but fails on new problems

Try It Yourself

Toggle between training and test data to see the difference. The overfit chef memorized every training recipe but can't handle new orders. The well-trained chef generalizes and succeeds everywhere.

🔁 Epochs = Multiple Practice Runs

Training the model multiple times through the entire dataset

Try It Yourself

Press "Run Training" to loop through epochs. Each pass improves the pizza as the model learns. Watch both training and validation loss decrease. Too many epochs without early stopping can lead to overfitting.

🧪 Weights = Secret Recipe Parameters

Tunable knobs that control how the model makes decisions

Try It Yourself

Slide the ingredient sliders to discover the perfect recipe. Each weight (θ) controls how much that ingredient contributes. Negative weights suppress ingredients; positive ones enhance them. Train it to find the optimal values!

📝 Backpropagation = Chef Tracing Mistakes Backwards

Finding where errors originated and adjusting the recipe

Try It Yourself

Press "Spot the Mistake!" to see how errors propagate backwards through the network. The red wave shows which weights contributed most to the error. The chef corrects them, and forward pass succeeds.