A robust error suppressor function, which is independent of the underl
ying probability density function and convenient for computer programs
,is proposed to robustify the backpropagation learning algorithm of mu
ltilayer feedforward networks. Its performance is studied for simulate
d data sets.