Practicum 3

Key Word(s): Saliency Maps, Debug Models

Readings

• Sanity Checks for Saliency Maps
• The (Un) reliability of saliency methods
• Interpretation of Neural Networks is Fragile
• “How do I fool you?”: Manipulating User Trust via Misleading Black Box Explanations
• Explanations can be manipulated and geometry is to blame
• How can we fool LIME and SHAP? Adversarial Attacks on Post hoc Explanation Methods