Research map / Label-efficient and robust learning / Recognition and long-tail
Striking the Right Balance With Uncertainty
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 uncertainty-based margin enforcement instead of standard softmax (ablation) | CNN on imbalanced MNIST · MNIST imbalanced split | accuracy: 97.2 -> 98.3 | Helped |
| Add sample-level uncertainty modeling to uncertainty-based margins (ablation) | CNN + UMM on imbalanced MNIST · MNIST imbalanced split | accuracy: 98.3 -> 98.7 | Helped |
| Use dropout rate 0.3 rather than 0.5 (ablation) | CNN + UMM + SUM · MNIST imbalanced split | accuracy: 98.7 -> 98.3 | Hurt |
| Increase dropout rate from 0.5 to 0.7 (ablation) | CNN + UMM + SUM · MNIST imbalanced split | accuracy: 98.7 -> 98.7 | No clear effect |
| Set margin m=2 rather than m=3 (ablation) | Proposed method · CIFAR-10 imbalanced split | accuracy: 80.6% -> 80.2% | Hurt |
| Set margin m=4 rather than m=3 (ablation) | Proposed method · CIFAR-10 imbalanced split | accuracy: 80.6% -> 80.5% | Hurt |
| Use more than N=10 dropout samples for uncertainty estimates (ablation) | Proposed method · CIFAR-10 imbalanced split | accuracy: 80.6% -> 80.6%, 80.7% | No clear effect |
| Incorporate class and sample uncertainty in the classifier | This Paper · CIFAR-10 imbalanced split | F1: 75.0±12.3 -> 80.4±6.3 | Helped |
| Use uncertainty estimates to set the margins (ablation) | Additive Margin loss on CIFAR-10 · CIFAR-10 imbalanced split | accuracy: 78.3 -> 78.9 | Helped |
| Use uncertainty estimates to set the margins (ablation) | Arc Margin loss on CIFAR-10 · CIFAR-10 imbalanced split | accuracy: 78.1 -> 78.4 | Helped |
| Use uncertainty estimates to set the margins (ablation) | Large Margin loss on CIFAR-10 · CIFAR-10 imbalanced split | accuracy: 79.4 -> 79.9 | Helped |
| Use uncertainty estimates to set the margins (ablation) | Sphere Product loss on CIFAR-10 · CIFAR-10 imbalanced split | accuracy: 76.2 -> 77.5 | Helped |
Closest papers by meaning
- Balancing the Scales: A Theoretical and Algorithmic Framework for Learning from Imbalanced Data (ICML 2025 · similarity 0.88)
- Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 2024 · similarity 0.87)
- Class-Attribute Priors: Adapting Optimization to Heterogeneity and Fairness Objective (AAAI 2024 · similarity 0.86)
- Balanced Data, Imbalanced Spectra: Unveiling Class Disparities with Spectral Imbalance (ICML 2024 · similarity 0.86)
- A Data-Driven Measure of Relative Uncertainty for Misclassification Detection (ICLR 2024 · similarity 0.86)
- Leveraging Uncertainty Estimates To Improve Classifier Performance (ICLR 2024 · similarity 0.85)
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.