🕸️ The neural network visualizer
This mini-network must decide: “Is it a cat?” » 🐱 He receives 3 clues (the entries ), passes them through a layer of small calculators (the neurons ), and gives a verdict.
Each connection has a weight : the stronger it is, the more the signal matters. Learning means adjusting these weights. It's up to you to play with the sliders!
🎛️ Network weights (the importance of each index)
🐱 “Cat” score
🐶 “Not cat” score
🤔 The network hesitates…
💡 Challenge : adjust the weights by hand so that the network says “CAT” when ears + whiskers are turned up, and “NO CAT” as soon as there is barking. Then click “Autotrain”: the AI finds the setting on its own — that’s exactly what learning is. Tip: a weight can be negative (the clue plays AGAINST the cat).
⚠️ Real AI does the same… with billion of sliders, automatically adjusted millions of times.
👩🏫 The supervised learning simulator
Train your own AI to distinguish between animals 🐾 vehicles 🚗. Give him examples labeled , train her, then test her on things she doesn't have never seen before .
1️⃣ Training data (with labels)
2️⃣ The model
🤖Waiting for examples… (0 examples)
3️⃣ The exam: NEVER seen examples
💡 Experience to do: ① train with only 4 examples → test (often average). ② Add up to 16-20 examples → test (much better: more data = better learning ). ③ Check “low variation data”, reset and start again → the model misses dogs and planes: he's never seen anything like it! Diversity matters as much as quantity.
📉 The overfitting demo: the student who learns by heart
We train an AI and we follow two notes : his success on the exercises déjà vu (training) and on exercises kept secret (test). Drag the time slider…and watch the curves separate.
TRAINING score (the known)
Score on the TEST (unheard of)
🌱 Training has not started. Click ▶️!
💡 The million dollar question: When should you STOP training? Look for the top of the orange curve — the pros call it early termination . After this point, the AI no longer learns: it recite .
🧠 Human translation: the student who repeats the same exercises 50 times ends up knowing the answers by heart (green curve at the top)… and fails the test with new exercises (orange curve which dips).