Traditionally, we design two different models, one using a loss LID, the other one using loss function Lagc, for face identication and age estimation, respectively. But this would increase carbon consumption due to training. To make your system greener, you want to design a single model that can perform both face identication and age estimation tasks, given the vector :3: E Rd of a face image. 1). (1 pt) For this multi-task system, please write down the loss function, which contains a hyperparameter to control the relative importance of the two tasks. 2). (2 pt) Will the multi-task system have a higher age estimation accuracy than the traditional single-task age estimation model? Please explain your answer in a few sentences. Suppose the hyperparameter is optimally chosen. 3). (2 pt) Will your answer to the above question change if we replace face identication with mustache prediction (predicting whether a face has mustache)? We assume mustache labels (1 for existence, 1 for non-existence) are provided