Linear Regression

> A straight line has exactly two numbers to learn — a slope and an intercept — and gradient descent finds both of them the same way it trains a network with a billion parameters: one small step against the gradient at a time.