Consistency Models Quiz

Answers and explanations appear as you take the quiz.

1. What core problem with diffusion models do consistency models address?
   • Diffusion sampling is slow because it needs many iterative steps
   • Diffusion cannot generate color images
   • Diffusion uses too little memory
   • Diffusion only works on text
2. What is the 'self-consistency' property these models are trained to satisfy?
   • Every image must be the same color
   • Every point on a noise-to-data trajectory maps to the same clean endpoint
   • The model must agree with a human labeler
   • The output must equal the input noise
3. What boundary condition must the consistency function satisfy at time zero?
   • It must output pure noise
   • It must output a black image
   • It must act as the identity (return the input unchanged)
   • It must output the dataset average
4. How can a consistency model be created from an existing diffusion model?
   • By deleting half its layers at random
   • By converting it to a decision tree
   • By training it only on text data
   • By consistency distillation of the pretrained diffusion model
5. What technique stabilizes consistency training by providing learning targets?
   • A target network updated as an exponential moving average
   • Randomly deleting the loss function
   • Doubling the learning rate each step
   • Using only one training image
6. How many sampling steps did the Latent Consistency Models paper report for high-quality 768×768 image generation?
   • Hundreds of steps in every setting
   • Two to four sampling steps
   • Exactly one training epoch
   • No sampling or model evaluation
7. What are two training routes introduced for consistency models?
   • Only supervised image classification
   • Only adversarial training with a discriminator
   • Consistency distillation from a pretrained diffusion model and standalone consistency training
   • Token prediction and masked language modeling
8. Which zero-shot image-editing tasks did the consistency-model paper demonstrate?
   • Speech diarization and speaker verification
   • Machine translation and summarization
   • Database query planning and code completion
   • Inpainting, colorization and super-resolution