ml-recall

#epsilon-prediction

3 questions from Diffusion Models

Timed test

3 questions · 3 min · score and explanations at the end

Take a test

Q 1 / 3Score 0 / 0

Shuffle questions

Question 1: Single choice

mediumAI-checkedDiffusion Models

DDPM trains a network $\epsilon_\theta(x_t, t)$ with the simplified loss $\lVert \epsilon - \epsilon_\theta(x_t, t) \rVert_2^2$ . What is the network asked to output?

The noise $\epsilon$ that was added to $x_0$ to make $x_t$ 1
The clean image $x_0$ , from which $\epsilon$ is computed afterwards2
A fresh sample of $x_{t-1}$ drawn from the forward process3
The variance $\beta_t$ of the forward step at time $t$ 4

Keys: 1–9 select · Enter check / next