“Why Should I Trust You?” Explaining the Predictions of Any Classifier

Papers We Love, Nov 2017

Transcript

1. “Why Should I Trust You?” Explaining the Predictions of Any Classifier Papers We Love Nov 2017
3. Can I trust the prediction?
4. Can I trust the prediction?
5. Trust • Trusting a prediction: whether a user trusts an individual prediction sufficiently to take some action based on it. • Trusting a model: whether the user trusts a model to behave in reasonable ways if deployed.
6. Trust Prediction Model
7. LIME: Local Interpretable Model-agnostic Explanations Local: global model can be complicated and hard to approximate Interpretable: because human Model-agnostic: works on any black box Explanation: back to “trust”
8. Local fidelity
9. Pick predictions: Submodular pick (SP)