Research map / Label-efficient and robust learning / OOD and anomaly detection
Estimating Example Difficulty Using Variance of Gradients
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 VoG instead of MSP for OoD detection | WideResNet-28-10 · CIFAR-10 vs CIFAR-100 | AUROC: 80.9 -> 89 | Helped |
| Use VoG instead of MSP for OoD detection | WideResNet-28-10 · CIFAR-10 vs Gaussian | AUROC: 78.1 -> 88.2 | Helped |
| Use VoG instead of MSP for OoD detection | ResNet-18 · CIFAR-10 vs Gaussian | AUROC: 92.7 -> 85.1 | Hurt |
| Use VoG instead of MSP for OoD detection | ResNet-18 · CIFAR-10 vs iSUN | AUROC: 85.5 -> 92.3 | Helped |
| Use VoG instead of MSP for OoD detection | DenseNet-BC · CIFAR-10 vs Gaussian | AUROC: 95.8 -> 88.2 | Hurt |
| Use VoG instead of MSP for OoD detection | DenseNet-BC · CIFAR-10 vs CIFAR-100 | AUROC: 91.4 -> 93.1 | Helped |
| Use VoG instead of MSP for OoD detection | all three tested architectures · CIFAR-10 OoD datasets, averaged across five datasets | AUROC: mean gain of 2.62% | Helped |
| Use VoG rather than model confidence to rank examples | ImageNet classifier · ImageNet-O | share of images in VoG percentile: 24% -> 30% | Helped |
| Compute VoG using early-stage rather than late-stage checkpoints (ablation) | CIFAR-10, CIFAR-100, and ImageNet classifiers · Test-set error by VoG rank | error versus VoG ranking | Mixed |
| Use predicted labels instead of true labels to compute VoG (ablation) | ImageNet classifier · ImageNet test set | test error by VoG rank | No clear effect |
| Use the last 3 checkpoints rather than all 32 checkpoints (ablation) | ImageNet ResNet-50 VoG ranking · ImageNet | ranking stability | No clear effect |
Closest papers by meaning
- Towards Provably Unlearnable Examples via Bayes Error Optimization (AAAI 2026 · similarity 0.84)
- Show Me What You Don’t Know: Efficient Sampling from Invariant Sets for Model Validation (NeurIPS 2026 · similarity 0.83)
- Reproduce, Replicate, Reevaluate. The Long but Safe Way to Extend Machine Learning Methods (AAAI 2024 · similarity 0.82)
- The VOROS: Lifting ROC Curves to 3D to Summarize Unbalanced Classifier Performance (AAAI 2025 · similarity 0.82)
- Taming Hyperparameter Sensitivity in Data Attribution: Practical Selection Without Costly Retraining (NeurIPS 2025 · similarity 0.82)
- Is Cross-validation the Gold Standard to Estimate Out-of-sample Model Performance? (NeurIPS 2024 · 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.