This question concerns neural networks. Consider a dataset consisting of RGB images of size 2020 pixels and a multi-class label that can take on 4 different values. Labels are encoded using one-hot encoding, so that each label is represented by the values [1,0,0,0],[0,1,0,0],[0,0,1,0] and [0,0,0,1], respectively. a) The first architecture that we will explore is one consisting of just one single fullyconnected layer. i) Use a diagram to describe this architecture. ii) Determine the number of parameters to be trained. iii) Is a dataset consisting of 1000 images appropriate to train this network, or would you expect the network to either underfit or overfit? [15 marks] b) The second architecture is a convolutional neural network. i) The first hidden layer is a convolutional layer defined by 10 filters of dimensions 33D. Determine D and the number of feature maps produced by the first layer. ii) Assuming that each feature map in the first hidden layer has dimensions 20x 20 and the second layer is a 22 max-pooling layer, determine the number of feature maps in the sernnd laver and their dimensions