Diffusion Forward Noising

Problem

Apply the closed-form DDPM forward noising step and report the cumulative alpha bar.

The cumulative alpha bar is the product of (1 - beta) over betas, and x_t = sqrt(alpha_bar) * x0 + sqrt(1 - alpha_bar) * noise. Return [round(x_t, 4), round(alpha_bar, 4)].

Test cases

case 1

in: [1,[0.1],0]

expected: [0.9487,0.9]

case 2

in: [1,[0.1],1]

expected: [1.2649,0.9]

case 3

in: [2,[0.01,0.02,0.04],-1]

expected: [1.6682,0.9314]

case 4

in: [0.5,[0.5,0.5],0.5]

expected: [0.683,0.25]

Starter codePython

def diffusion_forward(x0, betas, noise):
    # Your code here
    pass

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