SGLD Noise Scale

Problem

Compute the SGLD noise standard deviation for a given step size.

Return sqrt(2 * lr * temperature), or 0.0 when lr or temperature is not positive. This is the amount of Gaussian noise added per coordinate.

Hint

The noise scale grows with the square root of the step size and temperature.

case 1

in: [0.1,1]

expected: 0.4472135954999579

case 2

in: [0.01,2]

expected: 0.2

case 3

in: [0,1]

expected: 0

case 4

in: [0.5,0.5]

expected: 0.7071067811865476

Starter codePython

def sgld_noise_scale(lr, temperature):
    # Your code here
    pass

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