MediumCalculusca-205

Reverse-Mode AD Graph Cost

Problem

For a fully connected network with layer sizes layers = [n0, n1, ..., nL], the forward pass performs sum(n_i * n_{i+1}) multiply-adds.

Reverse-mode AD costs about twice the forward pass in the backward sweep, so the total is 3 * sum(n_i * n_{i+1}). Return the total operation count.

Hint

Reverse mode costs roughly the same order as the forward pass, with about 2x the constant.

case 1

in: [[2,3]]

expected: 18

case 2

in: [[3,4,2]]

expected: 60

case 3

in: [[5]]

expected: 0

case 4

in: [[1,1,1]]

expected: 6

Starter codePython

def reverse_ad_cost(layers):
    # Your code here
    pass

4 test cases · run your code against them in the editor