Research map / Label-efficient and robust learning / Recognition and long-tail
SelectiveNet: A Deep Neural Network with an Integrated Reject Option
Read the paperCheck 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 |
|---|---|---|---|
| Jointly train prediction and rejection for a target coverage instead of applying confidence-based rejection to a full-coverage model | SelectiveNet · CIFAR-10, SVHN, and Cats vs. Dogs | selective risk (% error): 8.98; 9.05; 15.40 | Helped |
| Jointly train prediction and rejection for a target coverage instead of using MC-dropout confidence | SelectiveNet regression · Concrete Compressive Strength | selective risk (MSE): 14.23; 17.09; 13.18; 7.23 | Helped |
| Train SelectiveNet for coverage 0.95 rather than use a full-coverage confidence baseline | SVHN · SVHN, coverage 0.95 | selective risk (% error): 1.40$\pm$0.01 vs 1.40$\pm$0.05 | No clear effect |
| Add an auxiliary prediction head and its full-coverage prediction loss (ablation) | SelectiveNet · Selective prediction training | selective risk | Helped |
| Use a small auxiliary-loss weight so the selective loss dominates training (ablation) | SelectiveNet · Selective prediction training | selective risk | Hurt |
| Calibrate the selection threshold on an independent unlabeled validation set | SelectiveNet · CIFAR-10 coverage calibration | selective risk (% error) | No clear effect |
Closest papers by meaning
- Selective Classification for Deep Neural Networks (NeurIPS 2017 · similarity 0.93)
- Adaptive Neural Networks for Efficient Inference (ICML 2017 · similarity 0.86)
- Hierarchical Selective Classification (NeurIPS 2024 · similarity 0.86)
- Confidence-aware Contrastive Learning for Selective Classification (ICML 2024 · similarity 0.85)
- Initial Guessing Bias: How Untrained Networks Favor Some Classes (ICML 2024 · similarity 0.83)
- Rejection via Learning Density Ratios (NeurIPS 2024 · similarity 0.83)
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.