Consider a dataset.

$\backslash$table$[[,1,2,3,4],[$X$1,1\mathrm{.}2,1\mathrm{.}6,0\mathrm{.}5,0\mathrm{.}7],[$X$2,1\mathrm{.}4,0\mathrm{.}5,0\mathrm{.}5,0\mathrm{.}2],[Y,1,1,0,0]]$

Let the weight matrices $w1$ and $w2$ be as follows:

$w1=\left[\begin{array}{cc}0\mathrm{.}16& 0\mathrm{.}1\\ 0\mathrm{.}02& 0\mathrm{.}7\end{array}\right],w2=\left[\begin{array}{cc}0\mathrm{.}05& 0\mathrm{.}33\end{array}\right]$

$[3]$ Draw the network diagram.

$[5]$ Using the sigmoid function as a non$-$linear activation function, perform the forward propagation. Report the accuracy rate.

$[10]$ Then, perform the backward propagation. Report the updated $w1$ and $w2$ when the learning rate value $0\mathrm{.}1$ is used.

$[5]$ Reperform the forward propagation and predict the class.

$[2]$ Is the neural network algorithm efficient? Explain why. If you are going to use other than a neural network, which classification model would you prefer?