AI Paper Atlas

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

Neural Concept Verifier: Scaling Prover-Verifier Games via Concept Encodings

Berkant Turan, Suhrab Asadulla, David Steinmann, Kristian Kersting, Wolfgang Stammer, Sebastian Pokutta

ICML 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Self-explaining and visual explanation networks

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
move the prover-verifier game from pixel space to concept encodingsMAC / NCV · CIFAR-100completeness (accuracy %): 15.27 -> 83.32Improved
move the prover-verifier game from pixel space to concept encodingsMAC / NCV · ImageNet-1kcompleteness (accuracy %): 35.06 -> 67.04Improved
move the prover-verifier game from pixel space to concept encodingsMAC / NCV · CLEVR-Hans3completeness (accuracy %): 96.59 -> 98.92Improved
replace linear CBM head with NCV prover-verifier gameCBM · CLEVR-Hans7completeness (accuracy %): 89.12 -> 97.89Improved
replace linear CBM head with NCVCBM · COCOLogiccompleteness (accuracy %): 58.84 -> 75.42Improved
replace linear CBM head with NCVCBM · ImageNet-1kcompleteness (accuracy %): 68.59 -> 67.04Hurt
add NCV (PVG on concepts) in place of Pixel-MAC for soundnessMAC / NCV · CIFAR-100soundness (robustness %): 96.31 -> 99.99Improved
add NCV instead of non-linear CBM with 1% clean dataCBM (non-lin.) · CLEVR-Hans3 shortcutval-test gap (pp): 4.26 -> 2.26Improved
add NCV instead of non-linear CBM with no clean dataCBM (non-lin.) · CLEVR-Hans3 shortcutval-test gap (pp): 3.66 -> 5.23Hurt
add NCV instead of non-linear CBM with 5% clean dataCBM (non-lin.) · CLEVR-Hans3 shortcutval-test gap (pp): 2.28 -> 0.71Improved
add NCV instead of non-linear CBM with no clean dataCBM (non-lin.) · CLEVR-Hans7 shortcuttest accuracy w/o shortcut (%): 90.69 -> 92.23Improved

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