Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Accurate Layerwise Interpretable Competence Estimation
Vickram Rajendran, William LeVine
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 |
|---|---|---|---|
| Use ALICE instead of Trust Score for competence estimation | SVM (RBF) · DIGITS, cross-entropy | mAP: .346 \textpm{}.086 -> .989 \textpm{} .0078 | Improved |
| Use ALICE instead of Trust Score for competence estimation | MLP (5 iterations) · DIGITS, cross-entropy | mAP: .579 \textpm{} .12 -> .966 \textpm{} .047 | Improved |
| Use ALICE instead of Trust Score for competence estimation | ResNet50 (1 epoch) · CIFAR100, cross-entropy | mAP: .346 -> .723 | Improved |
| Add the in-distribution probability term p(D|x) (ablation) | ALICE · CIFAR10/SVHN 10/90, distributional uncertainty | mAP: .100 \raisebox{.2ex}{$\scriptstyle\pm$} 0.000 -> .868 \raisebox{.2ex}{$\scriptstyle\pm$} 0.014 | Improved |
| Add the conditional error indicator term p(E<δ|x,c_j) (ablation) | ALICE · CIFAR100, cross-entropy, VGG16 underfit | mAP: .0369 \textpm{} .0041 -> .913 \textpm{} .012 | Improved |
| Add the conditional error indicator term p(E<δ|x,c_j) (ablation) | ALICE · DIGITS, cross-entropy, MLP underfit | mAP: .0538 \textpm{} .031 -> .999 \textpm{} .0015 | Improved |
| Add the class-probability term p(c_j|x,D) (ablation) | ALICE · Synthetic class overlap, 0.50 overlap | mAP: .730 \textpm{} 0.0 -> .960 \textpm{} 0.0 | Improved |
| Use ALICE rather than Trust Score | VGG16 (5 epochs) · CIFAR100, 0-1 error | AP: .762 -> .756 | Hurt |
| Use ALICE rather than Trust Score | MLP (5 iterations) · DIGITS, 0-1 error | AP: .996 \textpm{} .0036 -> .984 \textpm{} .014 | Hurt |
Closest papers by meaning
- Unlearning Evaluation through Subset Statistical Independence (ICLR 2026 · similarity 0.83)
- Suitability Filter: A Statistical Framework for Classifier Evaluation in Real-World Deployment Settings (ICML 2025 · similarity 0.83)
- Quantifying Uncertainty in Error Consistency: Towards Reliable Behavioral Comparison of Classifiers (NeurIPS 2025 · similarity 0.82)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.82)
- Accuracy vs. Accuracy: Computational Tradeoffs Between Classification Rates and Utility (NeurIPS 2026 · similarity 0.82)
- Probabilistic Stability Guarantees for Feature Attributions (NeurIPS 2025 · similarity 0.82)
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.
