Cоnvоlutiоnаl neurаl network. Consider the following pаrtially completed program which defines a convolutional neural network. Unless specified, assume zero padding and stride . (a) (3 pts) Suppose we feed this model with an input image tensor of shape 32 x 32 x 3, i.e., an RGB image with height and width of 32 pixels. After the first Conv2d layer (line 7), what is the shape of the output feature map? Height = [a1], Width = [a2], # Channels = [a3]. (b) (2 pts) How many trainable parameters does this Conv2d layer (line 7) have? You may disregard the bias terms. [b] (c) (1 pt) How many trainable parameters does the first max-pooling layer (line 9) have? [c] (d) (1 pt) How many filters/kernels are used in the second Conv2D layer (line 11)? [d] (e) (3 pts) In line 15, nn.Flatten() reshapes the output of the second convolutional block into a vector for each image. The network then passes the vector to fully-connected layers. The code for the first fully-connected layer is missing (line 18). Complete the definition based on surrounding code. [e] (f) (1 pt) This network is originally design for a classification task with 10 classes. Assume the task later changes to 20 classes. Which line must be changed to reflect this? Line [f]