Let X e Rnd be a data matrix, consisting of n samples, cach of which has d features, and let y R" be a vector of outcomes. For example, each row of X could have information about a house on the market, like its area, number of floors, number of bathrooms/bedrooms, etc. and each entry of y could be the price of that house. We are interested in building a model that predicts house prices from the set of its features, as listed above. Suppose that domain knowledge tells us that the relationship between the features and outcomes is linear; ideally, there exists a set of parameters e Rd such that Xo = y. However, n is large and there is noise in the acquisition of X and y, so this system is overdetermined. Still, we wish to find the best linear approximation, ie. we want to find the θ that minimizes the loss L(θ)-lly-X613. Assuming X has full column rank, compute θ': arg mine L(θ) in terms of X and y 2*