Fooling LIME and SHAP: Adversarial Attacks on Post hoc Explanation Methods
| Autor: | Himabindu Lakkaraju, Dylan Slack, Emily Jia, Sophie Hilgard, Sameer Singh |
|---|---|
| Jazyk: | angličtina |
| Rok vydání: | 2019 |
| Předmět: |
FOS: Computer and information sciences
Systematic error Computer Science - Machine Learning Post hoc Computer science Computer Science - Artificial Intelligence Machine Learning (stat.ML) 02 engineering and technology Machine learning computer.software_genre Machine Learning (cs.LG) Domain (software engineering) 03 medical and health sciences Adversarial system Statistics - Machine Learning 020204 information systems 0202 electrical engineering electronic engineering information engineering 030304 developmental biology 0303 health sciences business.industry Artificial Intelligence (cs.AI) Artificial intelligence business Classifier (UML) computer Criminal justice |
| Zdroj: | AIES |
| Popis: | As machine learning black boxes are increasingly being deployed in domains such as healthcare and criminal justice, there is growing emphasis on building tools and techniques for explaining these black boxes in an interpretable manner. Such explanations are being leveraged by domain experts to diagnose systematic errors and underlying biases of black boxes. In this paper, we demonstrate that post hoc explanations techniques that rely on input perturbations, such as LIME and SHAP, are not reliable. Specifically, we propose a novel scaffolding technique that effectively hides the biases of any given classifier by allowing an adversarial entity to craft an arbitrary desired explanation. Our approach can be used to scaffold any biased classifier in such a way that its predictions on the input data distribution still remain biased, but the post hoc explanations of the scaffolded classifier look innocuous. Using extensive evaluation with multiple real-world datasets (including COMPAS), we demonstrate how extremely biased (racist) classifiers crafted by our framework can easily fool popular explanation techniques such as LIME and SHAP into generating innocuous explanations which do not reflect the underlying biases. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|