1. The input image matrix of a convolutional neural network is [1 O 2] and the lter (kernel) matrix is [1 1], which has a stride of 1. The resulting feature map is passed through a ReLU layer to yield a rectied feature map. No max pooling layer is used. The output (which is a row vector) is then flattened to become a column vector (this can be simply done by transposing this output, i.e.the row vector, to become a column vector). The attened vector is then passed through a layer with a 2 x 2 weight matrix W whose all elements are 0's and 1's. The activation functions of this layer are linear f(:r) = :1: and it bias terms are 0's. Determine W such that the input is classied in class 2, encoded as [0 IF. This network does not have a softmax layer after the linear layer, although the answer would not change if it had a softmax layer. Hint: this problem has more than one solution