Consider a fully connected autoencoder (each hidden node is connected to all inputs and all outputs) with 2 inputs and 2 outputs and one hidden layer, linear activation function (output input) in both hidden and output nodes. This network is trained to minimize sum of squared error between the input and reconstructed input. Input data is represented by the following matrix A, whose rows corresponding to training records and columns corresponding to input dimensions. A= 2 0 -3 1 1-3 02 If one hidden node is used, after convergence of training process, a) What will be 2x1 weight matrix between input and hidden nodes? b) What will be 1x2 weight matrix between hidden and output nodes? c) What will be total reconstruction error? d) What is the minimum number of hidden nodes needed to achieve zero reconstruction error