Research map / Optimization / Implicit bias and generalization
On Generalization Error Bounds of Noisy Gradient Methods for Non-Convex Learning
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 Bayes-Stability with a data-dependent gradient-norm bound instead of a global Lipschitz-constant bound | Noisy gradient methods, including SGLD · General non-convex learning | generalization error bound | Helped |
| Replace normal training labels with randomly assigned labels | GLD and SGLD · Normal versus randomly labeled data | generalization error | Hurt |
| Replace normal training labels with randomly assigned labels | GLD and SGLD · Normal versus randomly labeled data | generalization error bound | Hurt |
| Add ℓ2 regularization and use the time-dependent Log-Sobolev bound | Continuous Langevin Dynamics · Bounded objective with ℓ2 regularization | generalization error bound | Helped |
| Replace the global gradient-norm scale with empirical gradient norms along the optimization path | SGLD · General non-convex learning | generalization error bound | Helped |
| Train longer after the expected empirical gradient becomes small | Noisy gradient methods · Later training stages | generalization error bound | No clear effect |
| Relax the step-size constraint constant from 20 to 2 | SGLD generalization bound · General non-convex learning | bound constant: 8.12 to 84.4 | Hurt |
| Use GLD on the full MNIST dataset without label corruption | GLD · MNIST | generalization error bound | Helped |
Closest papers by meaning
- Generalization Bounds for Gradient Methods via Discrete and Continuous Prior (NeurIPS 2022 · similarity 0.95)
- Time-Independent Information-Theoretic Generalization Bounds for SGLD (NeurIPS 2023 · similarity 0.93)
- Sharper Generalization Bounds for Learning with Gradient-dominated Objective Functions (ICLR 2021 · similarity 0.93)
- Generalization of noisy SGD in unbounded non-convex settings (ICML 2025 · similarity 0.92)
- Exactly Tight Information-theoretic Generalization Bounds via Binary Jensen-Shannon Divergence (ICML 2025 · similarity 0.91)
- Algorithmic Stability Unleashed: Generalization Bounds with Unbounded Losses (ICML 2024 · similarity 0.90)
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.