Research map / Label-efficient and robust learning / Recognition and long-tail
Justification-Based Reliability in Machine Learning
Nurali Virani, Naresh Iyer, Zhaoyuan Yang
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 ε-ball support rather than softmax thresholding | Grid Stability FC epistemic classifier · Grid Stability, large-noise | A_IK: 0.433 -> 0.750 | Improved |
| Use ε-ball support rather than softmax thresholding | Italy Power Demand CNN epistemic classifier · Italy Power Demand, large-noise | A_IK: 0.712 -> 1.000 | Improved |
| Use ε-ball support rather than softmax thresholding | MNIST FC epistemic classifier · MNIST, adversarial noise | F_IK: 0.851 -> 0.000 | Improved |
| Use ε-ball support rather than softmax thresholding | GTSRB CNN epistemic classifier · GTSRB, adversarial noise | F_IK: 0.631 -> 0.140 | Improved |
| Use ε-ball support rather than k-NN support | SynCon CNN epistemic classifier · SynCon, adversarial noise | F_IK: 0.440 -> 0.343 | Mixed |
| Use ε-ball support rather than k-NN support | MNIST FC epistemic classifier · MNIST, adversarial noise | F_IK: 0.222 -> 0.000 | Improved |
| Use ε-ball support union k-NN support (H-2) instead of ε-ball support (ablation) | SynCon CNN epistemic classifier · SynCon, adversarial noise | A_IK: 0.942 -> 0.957 | Improved |
| Use ε-ball support union k-NN support (H-2) instead of ε-ball support (ablation) | GTSRB CNN epistemic classifier · GTSRB, adversarial noise | F_IK: 0.140 -> 0.102 | Hurt |
| Use H-1 fallback to k-NN when the ε-ball is empty instead of ε-ball support (ablation) | SynCon CNN epistemic classifier · SynCon, adversarial noise | A_IK: 0.942 -> 0.942 | No clear effect |
| Use support from a logit layer only instead of an intermediate layer plus logit (ablation) | SynCon CNN epistemic classifier · SynCon, adversarial noise | A_IK: 0.957 -> 0.699 | Hurt |
| Use support from a logit layer only instead of an intermediate layer plus logit (ablation) | MNIST FC epistemic classifier · MNIST, adversarial noise | A_IK: 0.000 -> 0.000 | No clear effect |
| Use support from a logit layer only instead of an intermediate layer plus logit (ablation) | GTSRB CNN epistemic classifier · GTSRB, adversarial noise | A_IK: 0.018 -> 0.028 | Improved |
Closest papers by meaning
- CLEAR: Calibrated Learning for Epistemic and Aleatoric Risk (ICLR 2026 · similarity 0.86)
- Rethinking Aleatoric and Epistemic Uncertainty (ICML 2025 · similarity 0.86)
- Is Epistemic Uncertainty Faithfully Represented by Evidential Deep Learning Methods? (ICML 2024 · similarity 0.86)
- Epistemic Uncertainty Quantification For Pre-Trained Neural Networks (CVPR 2024 · similarity 0.85)
- Are Uncertainty Quantification Capabilities of Evidential Deep Learning a Mirage? (NeurIPS 2024 · similarity 0.85)
- Quantifying Uncertainty in Error Consistency: Towards Reliable Behavioral Comparison of Classifiers (NeurIPS 2025 · 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.
