He Init Variance

Problem

Return the He/Kaiming normal initialization variance 2 / fan_in, which keeps activations well-scaled in ReLU networks.

The signature is he_variance(fan_in).

Hint

He variance is the square of the He standard deviation.

case 1

in: [2]

expected: 1

case 2

in: [100]

expected: 0.02

case 3

in: [128]

expected: 0.015625

case 4

in: [512]

expected: 0.00390625

Starter codePython

def he_variance(fan_in):
    # Your code here
    pass

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