Under “improving reward,” another option that seems promising is using models to iteratively improve your training data, like in Iterative Label Refinement.
Under “improving reward,” another option that seems promising is using models to iteratively improve your training data, like in Iterative Label Refinement.