Sure they can, many output probabilities: http://deeplearning.net/tutorial/mlp.html#mlp