AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

On Locality of Local Explanation Models

Sahra Ghalebikesabi, Lucile Ter-Minassian, Karla DiazOrdaz, Christopher C. Holmes

NeurIPS 2021 · Learning theory and algorithms › Prediction and decision losses · approach: Shapley value attribution · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use small kernel bandwidth for reference sampling (Neighbourhood SHAP) instead of KernelSHAPSHAP · adversarial OOD classifier testAUC of OOD classifierImproved
Use Neighbourhood SHAP instead of KernelSHAPSHAP · COMPAS adversarial black boxrelative attribution to raceImproved
Use smaller bandwidth in Neighbourhood SHAPNeighbourhood SHAP · four datasets with XGBoostMSE of local predictionImproved
Use small instead of large kernel widthNeighbourhood SHAP · bike datatemperature attribution signMixed
Use small instead of large kernel widthNeighbourhood SHAP · Boston dataLSTAT attribution sign: correlation -0.76 vs +0.15Mixed
Apply Neighbourhood SHAP to DeepSHAP instead of standard DeepSHAPDeepSHAP · MNIST digit 8attribution sparsityImproved
Apply Neighbourhood DeepSHAP instead of DeepSHAPDeepSHAP · MNIST digit 8change in log odds of predicting 3No clear effect
Apply smoothing (Smoothed DeepSHAP) instead of DeepSHAPDeepSHAP · MNIST digit 8change in log oddsHurt
Fewer features in data set (e.g. iris) for bandwidth comparisonNeighbourhood SHAP · four datasetsMSE gap between bandwidthsMixed
Increase bandwidth in Smoothed SHAPSmoothed SHAP · Adult dataattribution of age and educationMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy