Jensen-Shannon Divergence is a smoothed and important divergence measure of informaton theory. It is defined as:
where M = (P+M)/2
D(P||M) and D(Q||M) is Kullback-Leibler Divergence.
Features:
1.JSD(P||Q) in [0, 1]. When JSD = 0, P and Q is the same, JSD = 1, they are different.
2.JSP(P||Q) = JSP(Q||P), which is different from DK(P||Q) and DK(Q||P)