AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Post hoc Explanations may be Ineffective for Detecting Unknown Spurious Correlation

Julius Adebayo, Michael Muelly, Harold Abelson, Been Kim

ICLR 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Post-hoc explanation analysis · 21 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 feature attribution to inspect a known, visible Tag artifact on the aligned classFeature attribution methods · Tag, across Bone Age, Knee, and Dog BreedsK-SSDImproved
Attempt to detect an unknown spurious signal using explanations on normal inputs onlyFeature attribution, concept activation, and training point ranking · Unknown spurious signalsCCMNo clear effect
Apply feature attribution methods to the non-visible background blurFeature attribution methods · Blurspurious signal detection reliabilityHurt
Use Guided Backprop rather than other tested attribution methods for known spurious signalsFeature attribution methods · Tag and StripeK-SSDMixed
Present feature attributions for inputs containing a spurious artifact even when the model is normalFeature attribution methods · Tag and StripeFAM-GTHurt
Explicitly test the spurious concept using TCAVTCAV · Known visible Tag and Stripe signalsK-SSDImproved
Use TCAV to test the non-visible Blur conceptTCAV · BlurK-SSDHurt
Rely on rankings of known non-spurious concepts to reveal an unknown artifactTCAV · Unknown spurious signalsCCMNo clear effect
Use training point ranking to identify reliance without knowing which artifact to inspectInfluence-function training point ranking · Spurious signals unknown at test timespurious reliance detectionNo clear effect
Give participants explanations without prior knowledge of possible artifactsSmoothGrad, TCAV, and Influence · Blinded user studyrejecting defective modelsNo clear effect
Give participants prior knowledge of possible artifacts and TCAV explanationsTCAV · Not-blinded user studyrejecting defective modelsImproved

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