Research map / Adversarial robustness and security / Certified and empirical robustness
Towards Reliable Learning for High Stakes Applications
Jinyang Gao, Junjie Yao, Yingxia Shao
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 GALVE reliability R(x)*confidence instead of confidence alone | reject-option ResNet-56 · CIFAR10 | error rate at 30% selectivity: 4.20 -> 1.90 | Improved |
| use GALVE instead of confidence | reject-option ResNet-56 · SVHN | error rate at 30% selectivity: 1.05 -> 0.43 | Improved |
| use plain GAN discriminator instead of confidence | reject-option ResNet-56 · CIFAR10 | error rate at 30% selectivity: 4.20 -> 3.4 | Improved |
| use entropy instead of confidence | reject-option ResNet-56 · CIFAR10 | error rate at 30% selectivity: 4.20 -> 4.32 | Hurt |
| train R(x) on high-variance samples G(cz), c=1.1, instead of plain GAN discriminator (c=1.0) (ablation) | GALVE · CIFAR10 | error rate at 30% selectivity: 3.4 -> 1.90 | Improved |
| scale factor c=1.2 instead of 1.1 (ablation) | GALVE · CIFAR10 | error rate at 30% selectivity: 1.90 -> 2.67 | Hurt |
| scale factor c=0.9 (below 1) instead of c=1.0 (ablation) | GALVE · CIFAR10 | error rate at 30% selectivity: 3.4 -> 3.27 | Improved |
Closest papers by meaning
- Robustness and Generalization via Generative Adversarial Training (ICCV 2021 · similarity 0.86)
- On Predicting Generalization using GANs (ICLR 2022 · similarity 0.85)
- Why Adversarially Train Diffusion Models? (ICLR 2026 · similarity 0.84)
- Improved Diffusion-based Generative Model with Better Adversarial Robustness (ICLR 2025 · similarity 0.83)
- Towards Trustworthy Deep Learning (AAAI 2024 · similarity 0.83)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 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.
