Research map / Learning theory and algorithms / Kernels and regression theory
Improved Sample Complexities for Deep Neural Networks and Robust Classification via an All-Layer Margin
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 |
|---|---|---|---|
| Train with all-layer margin optimization instead of standard SGD | WRN16-10 · CIFAR-10 validation, Baseline | validation error: 4.15% -> 3.42% | Helped |
| Train with all-layer margin optimization instead of standard SGD | WRN16-10 · CIFAR-10 validation, No data augmentation | validation error: 9.59% -> 6.74% | Helped |
| Train with all-layer margin optimization instead of standard SGD | WRN16-10 · CIFAR-10 validation, 20% random labels | validation error: 9.43% -> 6.72% | Helped |
| Train with all-layer margin optimization instead of standard SGD | WRN28-10 · CIFAR-10 validation, Baseline | validation error: 3.82% -> 3.00% | Helped |
| Train with all-layer margin optimization instead of standard SGD | WRN28-10 · CIFAR-10 validation, No data augmentation | validation error: 8.28% -> 6.47% | Helped |
| Train with all-layer margin optimization instead of standard SGD | WRN28-10 · CIFAR-10 validation, 20% random labels | validation error: 8.17% -> 6.01% | Helped |
| Train with all-layer margin optimization instead of standard SGD | WRN16-10 · CIFAR-100 validation, Baseline | validation error: 20.12% -> 19.14% | Helped |
| Train with all-layer margin optimization instead of standard SGD | WRN16-10 · CIFAR-100 validation, No data augmentation | validation error: 31.94% -> 26.09% | Helped |
| Train with all-layer margin optimization instead of standard SGD | WRN28-10 · CIFAR-100 validation, Baseline | validation error: 18.85% -> 17.78% | Helped |
| Train with all-layer margin optimization instead of standard SGD | WRN28-10 · CIFAR-100 validation, No data augmentation | validation error: 30.04% -> 24.67% | Helped |
| Use robust AMO instead of standard robust training | WideResNet16-10 · CIFAR-10 robust validation, PGD 50 steps, 10 restarts, ℓ∞ ε=8 | robust validation error: 48.75% -> 45.94% | Helped |
| Use robust AMO instead of standard robust training | WideResNet28-10 · CIFAR-10 robust validation, PGD 50 steps, 10 restarts, ℓ∞ ε=8 | robust validation error: 45.47% -> 42.38% | Helped |
Closest papers by meaning
- Large Margin Deep Networks for Classification (NeurIPS 2018 · similarity 0.90)
- Predicting the Generalization Gap in Deep Networks with Margin Distributions (ICLR 2019 · similarity 0.89)
- Input Margins Can Predict Generalization Too (AAAI 2024 · similarity 0.88)
- MMA Training: Direct Input Space Margin Maximization through Adversarial Training (ICLR 2020 · similarity 0.88)
- Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers (NeurIPS 2024 · similarity 0.87)
- Sharper Guarantees for Learning Neural Network Classifiers with Gradient Methods (ICLR 2025 · 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.