Exercise 7.1.9

If $m=4$ and $N=6$ with 3 hidden layers.

• The first hidden layer, weight matrix is 6 by 4, and bias is 6 by 1, so we have total 30 parameters.
• The second hidden layer, weight matrix is 6 by 6, and bias is 6 by 1, so we have total 42 parameters.
• The third hidden layer, weight matrix is 6 by 6, and bias is 6 by 1, so we have total 42 parameters.
• The output layer, weight matrix is 1 by 6, no bias, so we have 6 parameters.

In total we have $30+42+42+6=120$ parameters.

There are 6 ReLU functions in each total, with a total of 18 ReLU functions in the network.