$1\mathrm{.}6$ The main difference between Stochastic Gradient Descent approach and RMSProp approach is:

a$)$ SGD works on the sign fluctuation in $w,\text{'}$s for a specific $(i,j)$ over past many steps while

RMSProp works on the magnitude variations in $w_i\text{'}$s for a specific $($i$,$ i$)$ over past many

steps

$(1)$ SGD works on the sign fluctuation in $w_j\text{'}$s for a specific $($i$,$ j$)$ over past many steps while

RMSProp works on the magnitude variations in $w_i^{\text{'}}\text{'}$s comparing over different $,)\text{'}$s over

past many steps

$-$c$)$ the value of $$ in RMSProp is kept at higher values compared to value of $$ in SGD

d$)$ SGD works on magnitude variations in $w$;$\text{'}$s for a specific $(i,j)$ over past many steps while

RMSProp works on sign fluctuations in $w,\text{'}$s for a specific $(1,j)$ over past many steps.

$1\mathrm{.}7$ The value of the $($output of$)$ bias neurons in ANNs are set at $+1$ by default. Suppose the bias neuron

in a particular layer is set at $-1.$ Then: