The neural network, opened all the way up
The stage animates every idea as it is spoken: the neuron summing and firing, the forward pass flowing through the layers with real numbers, the loss landscape, the backpropagation blame-wave glowing red as gradients are computed link by link, the optimizer stepping downhill, the two loss curves forking at overfitting, the CNN’s filter sliding, and the LSTM’s gates breathing. English karaoke subtitles.
Run the forward pass — then backprop one step
Drive the tiny 2-2-1 network’s inputs and watch the forward pass compute z, ReLU and σ live, exactly as the video does by hand. Then fire one backprop step and watch every gradient appear and the weights nudge downhill.
Fifteen cards, one concept each
The neuron, activations, the loss functions, backprop, the optimizers, initialization, regularization, and the two specialized architectures — what each is, and the trap that bites. Filter as you revise.
Every term of deep learning, searchable
From the weighted sum to the vanishing gradient — the working vocabulary of neural networks, each with its formula or origin and its one-line meaning.
| term | family | formula / origin | what it means |
|---|
The Module 5 mastery check
Ten questions drawn from a pool of nineteen, reshuffled on every load — answers and order both. Score 70%+ before Module 6 — the transformer is waiting.