Research map / Optimization / Implicit bias and generalization
Can Neural Nets Learn the Same Model Twice? Investigating Reproducibility and Double Descent From the Decision Boundary Perspective
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 |
|---|---|---|---|
| Use SGD + SAM instead of SGD | ResNet-18 · CIFAR-10 | reproducibility: 83.74% -> 87.22% | Helped |
| Use SGD + SAM instead of SGD | VGG · CIFAR-10 | reproducibility: 80.92% -> 84.21% | Helped |
| Use SGD + SAM instead of SGD | MLPMixer · CIFAR-10 | reproducibility: 66.51% -> 68.06% | Helped |
| Use SGD + SAM instead of SGD | ViT · CIFAR-10 | reproducibility: 75.13% -> 75.19% | Helped |
| Use SGD + SAM instead of SGD | MLPMixer · CIFAR-10 | test accuracy: 82.04 -> 82.18 | No clear effect |
| Use SGD + SAM instead of SGD | ViT · CIFAR-10 | test accuracy: 75.49 -> 74.72 | Hurt |
| Add mixup training (ablation) | ResNet-18 · CIFAR-10 | decision-boundary similarity: 0.774 | No clear effect |
| Add mixup training (ablation) | ViT · CIFAR-10 | decision-boundary similarity: 0.808 | No clear effect |
| Increase ResNet-18 width toward the interpolation threshold | ResNet-18 · CIFAR-10 with label noise | fragmentation | Mixed |
| Increase ResNet-18 width toward the interpolation threshold | ResNet-18 · CIFAR-10 with label noise | reproducibility | Mixed |
| Train with label noise | ResNet-18 near the interpolation threshold · CIFAR-10 | decision-region fragmentation | Hurt |
| Distill from a pretrained teacher rather than train the student in the standard fashion | Student models · CIFAR-10 decision regions | teacher-student reproducibility | Helped |
Closest papers by meaning
- GNNBoundary: Towards Explaining Graph Neural Networks through the Lens of Decision Boundaries (ICLR 2024 · similarity 0.87)
- When Do Models Actually Decide? Mapping the Layer-Wise Decision Timeline in Pretrained Neural Networks (CVPR 2026 · similarity 0.86)
- Mind the Boundary: Coreset Selection via Reconstructing the Decision Boundary (ICML 2024 · similarity 0.86)
- Simplicity Bias and Optimization Threshold in Two-Layer ReLU Networks (ICML 2025 · similarity 0.85)
- Unraveling the Enigma of Double Descent: An In-depth Analysis through the Lens of Learned Feature Space (ICLR 2024 · similarity 0.84)
- Understanding and Diagnosing Deep Reinforcement Learning (ICML 2024 · 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.