Initialize all $v_{ih}$ and $w_{hj}$ to rand$(\{$:$-0\mathrm{.}01,0\mathrm{.}01)$

Repeat

For all $(x^t,r^t)inx$ in random order

For $h=1,$dots,$H$

$z_h$larrsigmoid$(w_h^T{x}^t)$

For $i=1,$dots,$K$

$y_i=v_i^{T}z$

For $i=1,$dots,$K$

$v_i=(r_i^t-y_i^t)z$

For $h=1,$dots,$H$

$w_h=({\displaystyle \underset{i}{\overset{?}{}}}(r_i^t-y_i^t)v_{ih})z_h(1-z_h)x^t$

For $i=1,$dots,$K$

$v_{i}larr{v}_i+v_i$

For $h=1,$dots,$H$

$w_{h}larr{w}_h+w_h$

Until convergence

Figure $11\mathrm{.}11$ Backpropagation algorithm for training a multilayer perceptron

for regression with $K$ outputs. This code can easily be adapted for two$-$class

classification $($by setting a single sigmoid output$)$ and to $K>2$ classification $($by

using softmax outputs$).|$

Use Training Data $=[((0,0),0),((0,1),1),((1,0),1),((1,1),0)]$

Track Epochs: Number of times through all the data

Record accuracy and squared error at different points in training.

Provide:

Initial Random Weights

Learning Rate

Final Weights

Outputs for data

Squared Error

Accuracy

Provide learning experiment results for at least $5$ times with different initial random weights.

Also, provide learning experiment results for $5$ times with a batch version of algorithm.

Do all work in Colab Jupyter Notebook