Q$6$

Which is true about the regularization?

$($A$)$ If we add a regularization term $\backslash (\backslash |\backslash$theta$\backslash |\_\{4\}^\{4\}\backslash )($i$.$e$.,$ the $\backslash (\backslash$ell$\_\{4\}\backslash )$ norm of the parameters$)$ to linear regression, the optimal solution will shrink

$($B$)$ If we do early stopping, then we will get a solution with a larger norm

$($C$)$ Adding flipped images to a training dataset of images will lead to worse performance

$($D$)$ Adding more noise to the input data for training will lead to better performance

Solution: