#vc-bound

Question 1: Select all that apply

Structural risk minimization (SRM) selects among nested model classes by comparing VC-type upper bounds on the generalization error that hold with probability $1 - \delta$ over the training set. Which statements are true? Select all that apply.

The confidence parameter $\delta$ can change which model SRM selects1
SRM selects the model whose upper bound on the expected error is smallest2
SRM is guaranteed to find the model with the lowest expected error3
Shrinking the covariance estimates of a Gaussian classifier toward a fixed prior matrix lowers its VC dimension4

Timed test