Different
kinds of weights were tried in the regression process in which case the least squares problem
becomes the problem of minimizing
{ }2102 ,...),()(log∑ −
i
iiii rMfPGXw ,
with respect to the parameters of the model f. The weights wi were generally of the form
),( rMq
p
wi = ,
where p was a normalization constant and usually
q(M,r) was a discrete density distribution
made to correct
/media/vedurstofan/utgafa/skyrslur/2009/VI_2009_012.pdf