Research map / Label-efficient and robust learning / Recognition and long-tail
Why ReLU Networks Yield High-Confidence Predictions Far Away From the Training Data and How to Mitigate the Problem
Check 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 |
|---|---|---|---|
| Adversarially train on noise inputs rather than only training on clean data | MNIST LeNet · MNIST test | test error: 0.51% -> 0.66% | No clear effect |
| Adversarially train on noise inputs rather than only training on clean data | SVHN ResNet · SVHN test | test error: 3.53% -> 3.52% | No clear effect |
| Adversarially train on noise inputs rather than only training on clean data | CIFAR-10 ResNet · CIFAR-10 test | test error: 8.87% -> 8.44% | Helped |
| Adversarially train on noise inputs rather than only training on clean data | CIFAR-100 ResNet · CIFAR-100 test | test error: 31.97% -> 32.24% | Hurt |
| Adversarially train on noise inputs to lower confidence | MNIST LeNet · MNIST-trained, Fashion-MNIST | MMC: 0.654 -> 0.239 | Helped |
| Adversarially train on noise inputs to lower confidence | SVHN ResNet · SVHN-trained, LSUN CR | MMC: 0.722 -> 0.148 | Helped |
| Adversarially train on noise inputs to lower confidence | CIFAR-10 ResNet · CIFAR-10-trained, SVHN | MMC: 0.800 -> 0.263 | Helped |
| Adversarially train on noise inputs to lower confidence | CIFAR-100 ResNet · CIFAR-100-trained, SVHN | MMC: 0.570 -> 0.234 | Helped |
| Add adversarial neighborhood optimization to noise confidence training | MNIST LeNet · MNIST-trained, adversarial noise | MMC: 0.102 -> 0.162 | Hurt |
| Add adversarial neighborhood optimization to noise confidence training | SVHN ResNet · SVHN-trained, adversarial noise | MMC: 0.946 -> 0.101 | Helped |
| Add adversarial neighborhood optimization to noise confidence training | CIFAR-10 ResNet · CIFAR-10-trained, adversarial noise | MMC: 0.985 -> 0.112 | Helped |
| Add adversarial neighborhood optimization to noise confidence training | CIFAR-100 ResNet · CIFAR-100-trained, adversarial noise | MMC: 0.985 -> 0.013 | Helped |
Closest papers by meaning
- Formal Guarantees on the Robustness of a Classifier against Adversarial Manipulation (NeurIPS 2017 · similarity 0.88)
- Towards neural networks that provably know when they don't know (ICLR 2020 · similarity 0.88)
- Training for Faster Adversarial Robustness Verification via Inducing ReLU Stability (ICLR 2019 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.87)
- Benign overfitting in leaky ReLU networks with moderate input dimension (NeurIPS 2024 · similarity 0.86)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.86)
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.