1. What is the risk with tuning hyperparaneters using a test dota set? 2. You design a fally connected neural network architecture where all activation are m moide. You initialine the weighiss with large positive numbers. Is this in good hilen? Rixplais youit answer. 3. You are given n dataret of 1010 grayscale insegs. Your frot is to beild a 5 class clansifier. Yon hase to adopt one of the following two options (a) the input is flatteaed into a too-dimenalonal vector, followed by a fully-connected layer with if nemrone (b) the input le directly given to a comblutional layer with five 1010 fitters Can the two approaches achlove similar training renulta? If so, which one you wonld chorm and wily? 4. In a CNN, the shape of input is (n,n,nw,nc)=(10,10,1). There are five 44 convolutionial filters with 'valid' paddine (no padding) and a stride of (2,2). What is the output shape after performing the convolution atep in Figure i7 Write your answer in the following format: (n,nw,nc) 5. Consider an input image of shape 5005003. You flatten this image and tse a fally consected layer with I00 hidder units. (a) What is the shape of the weight matrix of this layer? (b) What is the shape of the corresponding blas vector? 6. Consider an input image of shape 5005003. You run this image in a convolutional layer with 10 filters, of lectnel siar 55. How mazy parameters does this layer have? 7. Give a mothod to fight exploding gradient in fally-connected neural net works 8. Given an input volume of shape (10,10,3), you consider using otie of the two following layers: - Fully-connected Layer with 2 neurons, with biaces - Convolutional layer with throe 2x2 filters (with biasos) with 0 padding and a stride of 2 If you ase the fully-connected layer, the input volume is "fattened" into a column voctor belore being fed into the layer. What is the difference in the number of trainable parameters between these two layers? 9. Cite 3 layers commonly used in a convolutional neural network. 10. As you train your model, you realize that you do not have enough data. Cite 3 data augmentation techniques that can be used to overcome the shortage of data. 11. Although Pooling layers certainly cause a loss of information between Convolutional layers, why would we add Pooling layers to our network? 12. What is the purpose of using x convolution