Neural networks from a single neuron up to the mechanisms every modern architecture is built from.
Why deep learning works: stacking simple functions to approximate anything, and what that actually buys you over hand-written…
The single-neuron model that started it all: a weighted sum, a threshold, and a straight line through your data.
How a network turns a wrong guess into a slightly better set of weights, including the vanishing gradient problem that shapes…
Sigmoid through Swish, compared side by side and playable: drag the input and watch every curve respond.
The number training is always trying to shrink: MSE, MAE, Huber for regression, cross-entropy for classification.
The expand, bend, contract pattern behind every dense layer, and where most of a network's parameters actually live.
Beyond plain gradient descent: momentum, Adam, and how modern training actually takes each step.