Let us now apply MWM to an online learning problem like the perceptron. Suppose we are

given some examples $($x$1,$ y$1),...,($xt

$,$ yt$),...$ in an online manner where xi in $[1,1]$d and

yi in $\{1,1\}.$ We know that yi $=$ sign$($hw$,$ xii$)$ where w$,$ the unknown true coeffecient

vector satisfies