MIT and IBM Research have developed saliency cards to help users choose the best method for evaluating machine-learning models.
The cards provide standardized documentation of saliency methods, including strengths, weaknesses, and explanations to help users interpret them correctly.