For all three algorithms, the computation starts with an initialization of the center positions and
followed by iterative refinement of these positions. Many experimental results show that KHM is
essentially insensitive to the initialization of the centers than KM and EM. The dependency of the
K-Means performance on the initialization of the centers is a major problem; a similar issue exists
for an alternative algorithm, Expectation Maximization(EM), although to a lesser extent. Many
papers have been published to find good initializations for KM [BF98]. This paper takes a totally
different approach by changing MIN() used in KM to HA() (Harmonic Average), which is similar
to MIN() but “softer”, to make the performance function “easier to optimize” by an algorithm that
is essentially insensitive to initialization. More explanation will be given later.