Learn2Agree: Fitting with Multiple Annotators without Objective Ground Truth