3. Regularization (4 pts.): We have mentioned how when we are looking for parameters that optimize a loss/error function, sometimes we want to ensure these parameters are not too big or too complex in order to avoid overfitting. This is called regularization. We will explore optimizing a regularized loss function here. Let {Z1,it ~ N(-1,.5), {Z2, ~ N(1, .5), and {} ~ N(0, .1). Generate N = 300 values of y, defined as: Yi = Biznit Bizzi + ci, where BY = 1 and ; = .1. In the following, you will only use {Yi, Z1,i, Z2{} and will assume no knowledge of Bi and By. Define your loss function as a regularized version of sum of squares: N 1 L(B) = N (Yi - BIZ1,i - B2Z2,i)2 + AlBil + AlBel. In this case we are not just trying to minimize the sum of squares but, in doing so, we are trying to keep the absolute value of the As small