论文笔记understanding black-box predictions via influence functions

Purpose **To formalize the impact of a training point on a prediction, we ask the counterfactual: what would happen if we did not have this training point, or if the values of this training point were

>>阅读原文<<