Research map / Optimization / Implicit bias and generalization
Asymmetric Valleys: Beyond Sharp and Flat Local Minima
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 |
|---|---|---|---|
| Apply SWA weight averaging instead of standard SGD | ResNet-110 · CIFAR-100 | test accuracy: 78.29% -> 78.94% | Helped |
| Apply SWA weight averaging instead of standard SGD | ResNet-164 · CIFAR-100 | test accuracy: 76.56% -> 80.69% | Helped |
| Apply SWA weight averaging instead of standard SGD | DenseNet-100 · CIFAR-100 | test accuracy: 71.46% -> 72.29% | Helped |
| Apply SWA weight averaging instead of standard SGD | ResNet-110 · CIFAR-100 | training accuracy: 97.52% -> 94.98% | Hurt |
| Apply SWA weight averaging instead of standard SGD | ResNet-164 · CIFAR-100 | training accuracy: 99.12% -> 97.48% | Hurt |
| Apply SWA weight averaging instead of standard SGD | DenseNet-100 · CIFAR-100 | training accuracy: 99.87% -> 99.84% | Hurt |
| Continue training from batch size 128 to batch size 32 | PreResNet-164 · CIFAR-100 | test accuracy | Helped |
| Average BN parameters rather than an equal amount of non-BN parameters (ablation) | ResNet-110 · CIFAR-10 | test accuracy | Helped |
| Average only BN parameters rather than all weights (ablation) | ResNet-110 · CIFAR-10 | test accuracy | Hurt |
| Add BN layers before applying weight averaging (ablation) | ResNet-8 · CIFAR-10 | test accuracy | Helped |
Closest papers by meaning
- Exploring and Exploiting the Asymmetric Valley of Deep Neural Networks (NeurIPS 2024 · similarity 0.90)
- Symmetries, Flat Minima, and the Conserved Quantities of Gradient Flow (ICLR 2023 · similarity 0.89)
- The Loss Surface of Deep and Wide Neural Networks (ICML 2017 · similarity 0.89)
- Exploring Landscapes for Better Minima along Valleys (NeurIPS 2025 · similarity 0.88)
- Implicit bias of SGD in $L_2$-regularized linear DNNs: One-way jumps from high to low rank (ICLR 2024 · similarity 0.87)
- No Wrong Turns: The Simple Geometry Of Neural Networks Optimization Paths (ICML 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.