AI Paper Atlas

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

Counterfactual Explanations Can Be Manipulated

Dylan Z Slack, Sophie Hilgard, Himabindu Lakkaraju, Sameer Singh

NeurIPS 2021 · Learning theory and algorithms › Prediction and decision losses · approach: Counterfactual explanations and recourse · 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
add learned perturbation delta before counterfactual searchWachter et al. on manipulated model · Communities and Crimecost reduction: 20.1xImproved
add learned perturbation delta before counterfactual searchDiCE on manipulated model · Communities and Crimerecourse cost: 42.63 -> 9.57Improved
add learned perturbation delta before counterfactual searchSparse Wachter on manipulated model · Communities and Crimerecourse cost: 52.05 -> 22.59Improved
add learned perturbation delta before counterfactual searchPrototypes on manipulated model · German Creditrecourse cost: 13.98 -> 4.69Improved
add delta to non-protected group on unmodified modelDiCE on unmodified model · Communities and Crimerecourse cost: 11.90 -> 11.50No clear effect
train manipulated model instead of unmodified modelWachter et al. · Communities and Crimerecourse cost disparity: 3.59 -> 0.37Improved
train manipulated model instead of unmodified modelDiCE · Communities and Crimerecourse cost disparity: 24.41 -> 6.99Improved
train manipulated model instead of unmodified modelWachter et al. · Communities and Crimetest accuracy: 81.2 -> 80.9No clear effect
train manipulated model instead of unmodified modelSparse Wachter · Communities and Crimetest accuracy: 81.2 -> 77.9Hurt
move from Communities and Crime (99 features) to German Credit (7 features)manipulated models · both datasetscost reduction: 20.1x -> 1.8xHurt
round delta for categorical features instead of numerical-only delta (ablation)Prototypes manipulated model · German Creditcost reduction: 2.3x -> 0.92xHurt
initialize counterfactual search at x plus N(0,1) noise (ablation)Wachter and DiCE counterfactual search · Communities and Crimecost reductionHurt

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