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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| move the prover-verifier game from pixel space to concept encodings | MAC / NCV · CIFAR-100 | completeness (accuracy %): 15.27 -> 83.32 | Improved |
| move the prover-verifier game from pixel space to concept encodings | MAC / NCV · ImageNet-1k | completeness (accuracy %): 35.06 -> 67.04 | Improved |
| move the prover-verifier game from pixel space to concept encodings | MAC / NCV · CLEVR-Hans3 | completeness (accuracy %): 96.59 -> 98.92 | Improved |
| replace linear CBM head with NCV prover-verifier game | CBM · CLEVR-Hans7 | completeness (accuracy %): 89.12 -> 97.89 | Improved |
| replace linear CBM head with NCV | CBM · COCOLogic | completeness (accuracy %): 58.84 -> 75.42 | Improved |
| replace linear CBM head with NCV | CBM · ImageNet-1k | completeness (accuracy %): 68.59 -> 67.04 | Hurt |
| add NCV (PVG on concepts) in place of Pixel-MAC for soundness | MAC / NCV · CIFAR-100 | soundness (robustness %): 96.31 -> 99.99 | Improved |
| add NCV instead of non-linear CBM with 1% clean data | CBM (non-lin.) · CLEVR-Hans3 shortcut | val-test gap (pp): 4.26 -> 2.26 | Improved |
| add NCV instead of non-linear CBM with no clean data | CBM (non-lin.) · CLEVR-Hans3 shortcut | val-test gap (pp): 3.66 -> 5.23 | Hurt |
| add NCV instead of non-linear CBM with 5% clean data | CBM (non-lin.) · CLEVR-Hans3 shortcut | val-test gap (pp): 2.28 -> 0.71 | Improved |
| add NCV instead of non-linear CBM with no clean data | CBM (non-lin.) · CLEVR-Hans7 shortcut | test accuracy w/o shortcut (%): 90.69 -> 92.23 | Improved |
Closest papers by meaning
- Generative Verifiers: Reward Modeling as Next-Token Prediction (ICLR 2025 · similarity 0.85)
- Verification Learning: Make Unsupervised Neuro-Symbolic System Feasible (ICML 2025 · similarity 0.85)
- Concise and Logically Consistent Conformal Sets for Neuro-Symbolic Concept-Based Models (NeurIPS 2026 · similarity 0.85)
- FastCAV: Efficient Computation of Concept Activation Vectors for Explaining Deep Neural Networks (ICML 2025 · similarity 0.85)
- Predictive Concept Decoders: Training Scalable End-to-End Interpretability Assistants (NeurIPS 2026 · similarity 0.84)
- Concept-RuleNet: Grounded Multi-Agent Neurosymbolic Reasoning in Vision Language Models (AAAI 2026 · similarity 0.84)
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.
