Research map / Optimization / Implicit bias and generalization
Label Noise SGD Provably Prefers Flat Global Minimizers
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 label-noise SGD from a poor full-batch global minimizer | ResNet18 · CIFAR10 test set, full-batch initialization | accuracy: 76% initialization; reaches the SGD baseline | Helped |
| Apply label-noise SGD from a poor adversarial global minimizer | ResNet18 · CIFAR10 test set, adversarial initialization | accuracy: 48% initialization; within 1% of the baseline | Helped |
| Use label-noise SGD instead of minibatch SGD at a poor global minimizer | ResNet18 · CIFAR10 test set, zero-training-loss initializations | test accuracy | Helped |
| Increase the learning rate | SGD with label noise · CIFAR10 | tr ∇²L | Mixed |
Closest papers by meaning
- Noise against noise: stochastic label noise helps combat inherent label noise (ICLR 2021 · similarity 0.95)
- On the Learning Dynamics of Two-layer Linear Networks with Label Noise SGD (AAAI 2026 · similarity 0.92)
- On the Noisy Gradient Descent that Generalizes as SGD (ICML 2020 · similarity 0.91)
- Bad Global Minima Exist and SGD Can Reach Them (NeurIPS 2020 · similarity 0.91)
- The alignment property of SGD noise and how it helps select flat minima: A stability analysis (NeurIPS 2022 · similarity 0.90)
- How Does Label Noise Gradient Descent Improve Generalization in the Low SNR Regime? (NeurIPS 2025 · 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.