Explanation-based Training with Differentiable Insertion/Deletion Metric-aware Regularizers
The author proposes an optimization method, ID-ExpO, that enhances the faithfulness of explanations by training machine learning predictors with insertion and deletion metric-aware regularizers.
