Write T or F before each question. 1. A loss function is used to penalize errors in prediction. _2. According to the NFL theorem, the expected prediction error is irrelevant to any learning algorithm. _3. In AdaBoost, if one training example is misclassified in the current iteration, its weight should be decreased in the next iteration. 24. In the generalized linear model y = g '(w'x+b), the link function g must be monotone and differentiable. _5. The under-fitting occurs when the training error rate is low but the testing error rate is high