QUESTION 10 A training set D for a classification task contains 15 p-dimensional feature vectors on and corresponding class labels yn {1,2,3} = {cat, dog, bird). To learn the labels you perform softmax regression with a cross- entropy loss function, which for the 15 points and a certain set of weights gives a value of H. You realise you had left out a 16th training point ($16,716) whose class label 716 is "bird". Instead of recomputing the cross- entropy for the 16 points, you decide to compute a function that depends only on the predicted probabilities (01,92,D3) last data point. The cross-entropy for all 16 training points is O a. 3 Hi-Eng; i = 1 Ob. H15+ InD3 OCH15 - InD3 O d. None of the above 3 Oe H15+ Zindi i=1