The difference between predicted and expected values when training a machine learning model.  Also called a "loss function," the cost function is not about dollars.  It is the difference between the prediction and the correct answer.  See 
AI backpropagation.