Derive a gradient descent based learning rule for a perceptron that uses

tanh as activation function. Assume that the perceptron has D inputs and a

bias term and we have N training examples $($Xi$,$ Yi$)$ I $=1,2,...,$ N to learn

weights of the perceptron. Please note that the derivative of tanh$($x$)$ is $1$

tanh$2($x$)$

$2.$ Show the mathematical working of Artificial Neural Network by taking

the case in figure below. First two columns are the input values for X$1$ and

X$2$ and the third column is the desired output. Show only $3$ iterations.

Learning rate $=0\mathrm{.}2$

Threshold $=0\mathrm{.}5$

Actual output $=$ W$1$X$1+$W$2$X$2$

Next weight adjustment $=$ Wn$+$ Wn

Change in weight $=$ Wn $=$ learning rate $*($desired output$-$ actual output$)$

$*$ Xn

Show two complete iterations for acquiring the desired output.

X$1$ X$2$ W$1$ W$2$ D Y W$1$ W$2$