Research map / Optimization / Implicit bias and generalization
A Mean Field Analysis Of Deep ResNet And Beyond: Towards Provably Optimization Via Overparameterization From Depth
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 |
|---|---|---|---|
| Use mean-field training instead of vanilla SGD | ResNet20 · CIFAR10 | error rate: 8.75 -> 8.19 | Helped |
| Use mean-field training instead of vanilla SGD | ResNet32 · CIFAR10 | error rate: 7.51 -> 7.15 | Helped |
| Use mean-field training instead of vanilla SGD | ResNet44 · CIFAR10 | error rate: 7.17 -> 6.91 | Helped |
| Use mean-field training instead of vanilla SGD | ResNet56 · CIFAR10 | error rate: 6.97 -> 6.72 | Helped |
| Use mean-field training instead of vanilla SGD | ResNet110 · CIFAR10 | error rate: 6.37 -> 6.10 | Helped |
| Use mean-field training instead of vanilla SGD | ResNet164 · CIFAR10 | error rate: 5.46 -> 5.19 | Helped |
| Use mean-field training instead of vanilla SGD | ResNeXt29(8×64d) · CIFAR100 | error rate: 17.92 -> 17.53 | Helped |
| Use mean-field training instead of vanilla SGD | ResNeXt29(16×64d) · CIFAR100 | error rate: 17.65 -> 16.81 | Helped |
Closest papers by meaning
- Generalization of Scaled Deep ResNets in the Mean-Field Regime (ICLR 2024 · similarity 0.91)
- Depthwise Hyperparameter Transfer in Residual Networks: Dynamics and Scaling Limit (ICLR 2024 · similarity 0.90)
- On global convergence of ResNets: From finite to infinite width using linear parameterization (NeurIPS 2022 · similarity 0.90)
- Neural Collapse is Globally Optimal in Deep Regularized ResNets and Transformers (NeurIPS 2025 · similarity 0.89)
- Deep Linear Network Training Dynamics from Random Initialization: Data, Width, Depth, and Hyperparameter Transfer (ICML 2025 · similarity 0.89)
- Feature Learning Dynamics in Infinite-Depth Neural Networks (NeurIPS 2026 · similarity 0.88)
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.