$*$NOTE: The implementation of each exercise should be in different notebooks. The confusion

matrix and test accuracy for each of the activation functions should be inside each notebook but

in different cells. The same for the learning rate and optimum outputs.

Implementation $1)$ Using Collab, design and train a Multi$-$layer Perceptron $($MLP$)$ with $1$ hidden

layer and using Backpropagation $($BP$)$ algorithm on the Breast Cancer Wisconsin Dataset $.$ Use a

splitting of $70/30,100$ epochs. Show in the code the confusion matrix and test accuracy for each

of the following activation functions:

a$)$ Identity Activation Function

b$)$ Sigmoid A$.$F

c$)$ Step A$.$F

d$)$ Tanh A$.$F

e$)$ ReLu A$.$F

Implementation $2)$ In a new notebook, repeat the same process but now with $3$ hidden layers.

Implementation $3)$ In another notebook, using the architecture you built in implementation $2$

and the best performing activation function, re$-$train the model by adding $2$ new parameters,

learning rate and momentum. The range of the values you should try $($between $5$ and $10$ trials$)$

is:

Learning rate: $0<$ $1$

Momentum: $0<$ $<1$

Print on the code, for each combination, the corresponding accuracy. What are the optimum

learning rate and momentum values?