Absolute error instead of Mean Square Error - NN

It is possible to change the MSE performance to a simple Error performance? So, considering the minimization of the (maximum) error in a neural network?

Thanks in advance!