a) For a fully connected multi-layer perceptron containing two hidden layers with 100 hidden units each which is designed for the MNIST classification problem, calculate the number of learnable parameters (i.e. parameters which are learned using the backpropagation algorithm). b) The figure shows the architecture of the famous LeNet-5 convolutional network. Calculate the number of trainable parameters and the number of connections (synaptic weights plus biases, i.e. in augmented space) Cx: Convolutional layer x, Sx: Subsampling layer x, Fx: Fully connected layer x C3.t maps 16910x10 C1:feature maps 602828 S4.L maps 16@55 INPUT 3232 S2 t 6014x1 C5:layer F6 aye 20 F6 layer OUTPUT 84 Ful cannectionGaussian connections Convolutions Full connection d) How do these numbers change in the ZF net? a) For a fully connected multi-layer perceptron containing two hidden layers with 100 hidden units each which is designed for the MNIST classification problem, calculate the number of learnable parameters (i.e. parameters which are learned using the backpropagation algorithm). b) The figure shows the architecture of the famous LeNet-5 convolutional network. Calculate the number of trainable parameters and the number of connections (synaptic weights plus biases, i.e. in augmented space) Cx: Convolutional layer x, Sx: Subsampling layer x, Fx: Fully connected layer x C3.t maps 16910x10 C1:feature maps 602828 S4.L maps 16@55 INPUT 3232 S2 t 6014x1 C5:layer F6 aye 20 F6 layer OUTPUT 84 Ful cannectionGaussian connections Convolutions Full connection d) How do these numbers change in the ZF net