Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Bias-Reduced Uncertainty Estimation for Deep Neural Classifiers
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 |
|---|---|---|---|
| Average scores from 30 evenly spaced snapshots beginning at 0.4T | Softmax Response · CIFAR-10 | E-AURC: 4.78 -> 4.49 | Helped |
| Average scores from 30 evenly spaced snapshots beginning at 0.4T | Softmax Response · CIFAR-100 | E-AURC: 50.97 -> 39.90 | Helped |
| Average scores from 30 evenly spaced snapshots beginning at 0.4T | Softmax Response · SVHN | E-AURC: 4.24 -> 3.77 | Helped |
| Average scores from 50 evenly spaced snapshots beginning at 0.4T | Softmax Response · ImageNet | E-AURC: 99.68 -> 94.77 | Helped |
| Average scores from 30 evenly spaced snapshots beginning at 0.4T | NN-distance · CIFAR-10 | E-AURC: 35.10 -> 4.70 | Helped |
| Average scores from 30 evenly spaced snapshots beginning at 0.4T | MC-dropout · CIFAR-10 | E-AURC: 5.03 -> 4.99 | No clear effect |
| Average scores from 30 evenly spaced snapshots beginning at 0.4T | Ensemble · CIFAR-100 | E-AURC: 34.73 -> 30.72 | Helped |
| Average scores from 10 evenly spaced snapshots rather than 30 (ablation) | Ensemble · SVHN | E-AURC: 3.69 -> 3.51 | Helped |
| Use 10 snapshots instead of 30 (ablation) | AES on Softmax Response · CIFAR-10 | E-AURC: 4.78 -> 4.81 | Hurt |
| Use 50 snapshots instead of 30 (ablation) | AES on Softmax Response · CIFAR-100 | E-AURC: 50.97 -> 39.68 | Helped |
| Select a snapshot per confidence layer using labeled data (PES) | Softmax Response · CIFAR-10 | E-AURC: 4.6342 ± 0.07 -> 4.3543 ± 0.06 | Helped |
| Select a snapshot per confidence layer using labeled data (PES) | Softmax Response · CIFAR-100 | E-AURC: 51.3172 ± 0.43 -> 41.9579 ± 0.39 | Helped |
Closest papers by meaning
- Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles (NeurIPS 2017 · similarity 0.90)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.90)
- Uncertainty-Aware Deep Classifiers Using Generative Models (AAAI 2020 · similarity 0.89)
- Conservative Uncertainty Estimation By Fitting Prior Networks (ICLR 2020 · similarity 0.89)
- Modeling Uncertainty by Learning a Hierarchy of Deep Neural Connections (NeurIPS 2019 · similarity 0.89)
- One Step Closer to Unbiased Aleatoric Uncertainty Estimation (AAAI 2024 · similarity 0.87)
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.