Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Optimization-based Uncertainty Attribution Via Learning Informative Perturbations
Hanjing Wang, Bashirul Azam Biswas, Qiang Ji
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 |
|---|---|---|---|
| add SAM-guided mask parameterization (ablation) | Eq. (3) uncertainty attribution · C10 problematic region detection | IoU: 0.333 -> 0.648 | Improved |
| add learnable (blur) perturbation (ablation) | Eq. (3) + SAM · SVHN problematic region detection | IoU: 0.709 -> 0.765 | Improved |
| add Gumbel-sigmoid reparameterization of binary mask (ablation) | Eq. (3) + SAM + LP · SVHN problematic region detection | IoU: 0.765 -> 0.829 | Improved |
| add SAM segments to KernelSHAP | KernelSHAP · C10 problematic region detection | IoU: 0.324 -> 0.676 | Improved |
| optimization-based mask vs best baseline for faithfulness with gradient descent refinement | Ours vs KernelSHAP · C10/C100/SVHN avg URR | URR: 0.770 -> 0.884 | Improved |
| optimization-based mask vs Occlusion for blurring refinement | Ours vs Occlusion · C10/C100/SVHN avg URR, blurring | URR: 0.528 -> 0.743 | Improved |
| vary lambda (3 -> 2.5 or 2) and tau (0.1 -> 0.07 or 0.15) (ablation) | Ours · C10 problematic region detection | IoU: 0.713 -> 0.677 to 0.688 | No clear effect |
| refinement by gradient descent vs blurring for perturbation-based LIME/Occlusion | LIME and Occlusion · URR faithfulness | URR: Occlusion avg 0.587 (gradient) vs 0.528 (blurring) | Mixed |
| Ours vs UA-Backprop on PartImageNet ADA | Ours · PartImageNet detection | ADA: 0.695 -> 0.635 | No clear effect |
Closest papers by meaning
- Gradient-Based Uncertainty Attribution for Explainable Bayesian Deep Learning (CVPR 2023 · similarity 0.90)
- Adaptive Uncertainty Estimation via High-Dimensional Testing on Latent Representations (NeurIPS 2023 · similarity 0.86)
- Uncertainty Quantification for Deep Regression using Contextualised Normalizing Flows (NeurIPS 2025 · similarity 0.85)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.84)
- Benchmarking Uncertainty Disentanglement: Specialized Uncertainties for Specialized Tasks (NeurIPS 2024 · similarity 0.84)
- Pixel-level Certified Explanations via Randomized Smoothing (ICML 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.
