Give example problems (training situations), where these alternate design objectives give suboptimal solutions in the minimum classification error sense:

• maximum likelihood
• minimum perceptron loss
• minimum squared error
• maximum mutual information