Chris Lin, Ian Covert, Su-In Lee: On the Robustness of Removal-Based Feature Attributions. NeurIPS 2023