QUESTION THREE

a$)$ Assume a $3$ input perceptron with a bias input of $1($it outputs $1$ if net $>0,$ else $0),$ the learning rate $=1$ and initial weights all set at $0$:$w_i=(d-o)x_i.$ In this case, the net is the decision point or the activation function.

The training set is as follows:

$001$ the desired output is $0$

$111$ the desired output is $1$

$101$ the desired output is $1$

$011$ the desired output is $0$

Completely train this perceptron.

$[15]$

b$)$ State five tasks for which the neural networks approach is well suited include.

$[5]$

QUESTION FOUR

a$)$ Evaluate the role of a reward signal in the design of learning agents.

$[15]$

b$)$ Explain the importance of smoothing in the design of agents that reason under uncertainty and over time.

$[5]$

QUESTION FIVE

a$)$ Show that the probability expression $P(A,B|C)=P(A|C)P(B|C)$ is equivalent to the probability expression $P(A|B,C)=P(A|C).$

$[15]$

b$)$ Let assume you are given a biased coin that when tossed comes up heads $80$ per cent of the time. Calculate the information content of the message telling the outcome of tossing of the biased coin.

$[5]$