2 articles
How repeatedly stepping in the direction of steepest descent minimizes functions in millions of dimensions — and trains essentially every modern neural network.
The algorithm that squares the number of correct decimal places with every iteration.