Bayesian Methods Quiz
Five questions spanning the course — the posterior over weights, predictive uncertainty, Gaussian processes, and Bayesian optimization.
Exercise
What is the key output of Bayesian linear regression that ordinary least squares does not provide?
Need a hint?
A distribution, not just a point.
Exercise
Bayesian linear regression with a Gaussian prior of precision α and noise precision β has a posterior mean equal to which familiar estimator?
Need a hint?
An L2-penalized regression.
Exercise
In an RBF-kernel Gaussian process, what does the length-scale ℓ control?
Need a hint?
How fast does correlation between two outputs decay with input distance?
Exercise
A Gaussian process and a plain neural network are both fit to 8 data points. You query both far outside the data range. What is the qualitative difference?
Need a hint?
Which model signals that it doesn't know?
Exercise
Why does Bayesian optimization typically find good hyperparameters in fewer trials than random or grid search?
Need a hint?
Does it use the results of past trials?
Exercise
A Gaussian process reports a calibrated predictive variance (error bar) at every input. Which slot of the ML project loop does that most directly strengthen?
Need a hint?
It changes what you can say about how much to trust a prediction.
Exercise
Transfer test — a technique this course did NOT teach: gradient clipping caps the gradient norm to stop training from blowing up. Which slot does it modify?
Need a hint?
It changes how the parameter update is applied, not the model or the loss.