Research map / Optimization / Implicit bias and generalization
Generalization Bounds of Stochastic Gradient Descent for Wide and Deep Neural Networks
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 the NTRF reference class to characterize data classifiability | SGD-trained deep ReLU networks · NTRF reference class | expected 0-1 error bound: \tilde\cO(n^{-1/2}) | Helped |
| Apply the bound under the comparison paper's margin assumptions | SGD-trained deep ReLU networks · NTRF reference class, small-error setting | sample complexity: \tilde\cO(\epsilon^{-2}) versus \tilde\cO( \mathrm{poly}(2^L,\gamma^{-1}) \cdot \epsilon^{-4} ) | Helped |
| Include gradients from all network layers in the NTK reference | SGD-trained deep ReLU networks · NTK training data | expected 0-1 error bound | Helped |
| Allow the first-layer gradient features in the two-layer NTRF reference class | Two-layer SGD-trained ReLU networks · Two-layer reference function class | generalization bound | Helped |
Closest papers by meaning
- Generalizability of Neural Networks Minimizing Empirical Risk Based on Expressive Power (ICLR 2025 · similarity 0.90)
- Fantastic Generalization Measures are Nowhere to be Found (ICLR 2024 · similarity 0.90)
- Sharper Guarantees for Learning Neural Network Classifiers with Gradient Methods (ICLR 2025 · similarity 0.90)
- Generalization of Scaled Deep ResNets in the Mean-Field Regime (ICLR 2024 · similarity 0.90)
- On the Impacts of the Random Initialization in the Neural Tangent Kernel Theory (NeurIPS 2024 · similarity 0.89)
- Generalization Error of Graph Neural Networks in the Mean-field Regime (ICML 2024 · similarity 0.89)
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.