8. Which of the following atateniems age irue about trainies an HNN? Helect alf that npply. (d) All the abeve 9. Whica of thin following in an ipplication of ILNNY Wolect alt that apply. (a) Text mining (b) Image caption (c) That and Spoech analywia (d) All the abow 10. Which of the following is FAL.4B ahout epocts in a neural netwirk? (a) Number of epoche in the natmber no time the whole training date is shicws to the network. (b) Number of epoctin is a learnable parameter. (c) Time to complete oae epoch depends on the bateh alse (d) None of the above 11. Ridge regrosion usen what penalty on the model parameter a ? (a) a1 (b) 1 (c) (a122 (d) none of the abowe 12. After applying a regulhrizstioa penalty in linear resrossion, you find that some of the ceefliciente of w are arroes ost. Which of the following pechalties might hive bens ured? (a) 141 norm (b) 12 norm (c) aty of the above 13. Which of the following techniques does NOI prevent a model from overfiting?? (a) Data augmentation (b) Dropout (c) Rarly stoppins (d) None of the abover 14. Consider the data sets (XIrain, Virain ) and (Xtest, Yien ). You want to norralize your data before training gour model. Which of the follewing propositions are true? (Circle all that apply.) (a) The normalking mean and varlance cornputed an the training set, and uacd to train the model, shoeld be used to normalize test data. (b) Jint data ahould be normalired with its own mean asd varinnce before being fed to the network at test time because the test distribution might be different from the train distribution. (c) Normalizing the inpat impscts the lasdscape of the loes function. (d) In imaging, just like for atructured data, normalization conalsts in subtracting the mean froco the iepet asd multiplying the rasult by the standard deviation. 15. A 2-layer fully connected neural network is defined as follows All activations are sigmoids and your optienizer is stachaatie gradieat deacent. You initialiab all the weights and biasss to zero and forward propogate an input xRn in the network. What is the output y
