-
The chain rule at scale: from calc-101 to backprop through a million parameters
A neural network with a million trainable parameters sounds as though it should require a million separate calculus problems.
A neural network with a million trainable parameters sounds as though it should require a million separate calculus problems.