Bayesian Deep Learning Quiz

Answers and explanations appear as you take the quiz.

1. How does a Bayesian neural network treat its weights differently from a standard one?
   • It treats each weight as a probability distribution rather than a single fixed value
   • It uses integer weights instead of decimals
   • It removes weights entirely
   • It shares one weight across all layers
2. What can a Bayesian approach provide that a single fixed-weight network does not directly represent?
   • A guarantee that every confidence score is calibrated
   • A distribution of predictions that reflects uncertainty about model weights
   • A proof that the model will detect every unfamiliar input
   • A way to remove all noise from the data
3. How can epistemic uncertainty change as relevant, representative training data are added?
   • It is guaranteed to reach zero with any added data
   • It is identical to label noise
   • It may decrease under the chosen model and data assumptions
   • It always increases linearly
4. Why are approximations like variational inference or MCMC needed?
   • Because neural networks have no loss function
   • Because GPUs cannot do addition
   • Because weights must be integers
   • Because the exact posterior over millions of weights is computationally intractable
5. What does Monte Carlo dropout do to estimate uncertainty?
   • Keeps dropout active at test time and runs the network many times, using the spread of outputs
   • Turns dropout off and trains longer
   • Removes all neurons permanently
   • Replaces the loss with random noise
6. Variational inference reframes posterior estimation as what kind of problem?
   • A sorting problem
   • An optimization problem that minimizes KL divergence (via the ELBO)
   • A pure random sampling problem with no objective
   • A hardware design problem
7. How does Bayes’ rule relate a prior, likelihood, and posterior?
   • The posterior is the prior plus the likelihood, without normalization
   • It is the likelihood divided by the prior without normalization
   • The posterior is proportional to the prior times the likelihood
   • It is the fixed point estimate of the most likely weight vector
8. How does MCMC approximate a Bayesian neural-network posterior?
   • By training a single point estimate with dropout disabled
   • By ranking model outputs with a fixed rule
   • By minimizing an ELBO with a simpler variational family
   • By drawing samples through a Markov chain whose stationary distribution targets the posterior