AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Use-Case-Grounded Simulations for Explanation Evaluation

Valerie Chen, Nari Johnson, Nicholay Topin, Gregory Plumb, Ameet Talwalkar

NeurIPS 2022 · Large language models › Language, safety and interpretability · approach: Data and evaluation methodology · 7 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
Provide LIME explanations instead of no explanationSimEvals agent · Forward simulationagent test accuracy: 88.7% -> 100%Improved
Provide LIME explanations instead of AnchorsSimEvals agent · Forward simulationagent test accuracy: 93.7% -> 100%Improved
Provide LIME explanations instead of SHAP, GAM, or no explanationSimEvals agent · Counterfactual reasoningagent test accuracy: 64.9% -> 99.7%; 63.5% -> 99.7%; 60.3% -> 99.7%Improved
Provide LIME explanations instead of other explanation settingsHuman participants · Counterfactual reasoninghuman test accuracy: 41.4% ± 6.14% to 69.4% ± 12.4%; 45.7% ± 7.19% to 69.4% ± 12.4%; 48.6% ± 5.61% to 69.4% ± 12.4%Improved
Provide SHAP explanations instead of LIMESimEvals agent · Missing values bugagent test accuracy: 75.9% -> 100%Improved
Provide GAM explanations instead of LIMESimEvals agent · Missing values bugagent test accuracy: 75.9% -> 100%Improved
Provide SHAP or GAM explanations instead of LIME or model predictionsHuman participants · Missing values bughuman test accuracy: 48.0% -> 67.4%; 40.7% -> 67.4%; 48.0% -> 64.4%; 40.7% -> 64.4%Improved
Increase the number of data-points per observation from 1 to 1000 (ablation)SimEvals agent with SHAP · Missing values bugagent test accuracy: 63.2% -> 100%Improved
Increase the number of data-points per observation from 1 to 1000 (ablation)SimEvals agent with GAM · Missing values bugagent test accuracy: 64.8% -> 100%Improved
Increase the number of data-points per observation from 1 to 1000 (ablation)SimEvals agent with LIME · Missing values bugagent test accuracy: 55.2% -> 75.9%Improved
Increase the number of data-points per observation from 1 to 1000 (ablation)SimEvals agent with Model Prediction · Missing values bugagent test accuracy: 57.5% -> 67.3%Improved

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