Ouestion No $03$:

Consider the neural network architecture with two inputs $($x$1,$ x$2)$ three hidden

layer neurons and two output neurons $(Y1,Y2.$ Assume that each neuron has a

bias term set equal to $-1$ and each neuron uses the sigmoid activation function

given as

$f(x)=\frac{1}{1+e^{-x}}$

Input Layer Hidden Layer

Weights of all neurons as shown on the figure are:

Hidden Neurons

Output Neurons

For the training example with input $x=[1,1]$ and Target $Y=[0,1]$

Compute Output

Compute the neural network output where $=1$

Error Calculation and Back Propagation $-$ For each of the output neurons,

compute the error term $k$ used by the back propagation algorithm to update

the weights.

Compute the error term $h$ for each of the hidde $n$ neurons.

Weight Update

Use the errors computed above to compute the updated weights of the neural

network. Take the value of learning rate $=0\mathrm{.}5.$