Suppose a self$-$attention layer transforms an input feature map of $HW1$ dimensions into

an output feature map of $HW1$ dimensions, where each spatial feature $($of size $111)$

is affected by all other spatial features of the input. Explain then, how a fully connected layer

of shape where $N=HW1,$ which also lets each output dimension be affected by

all input dimensions, is different from this self$-$attention layer. $[15$ points$]$