Research map / Optimization / Implicit bias and generalization
Coherent Gradients: An Approach to Understanding Generalization in Gradient Descent-based Optimization
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 |
|---|---|---|---|
| Increase training-label noise | Vanilla SGD on MNIST · MNIST | steps to reach a given accuracy | Hurt |
| Train with pristine labels rather than corrupted labels | Vanilla SGD on noisy MNIST · MNIST | learning rate | Helped |
| Reduce the number of pristine examples by increasing label noise | Vanilla SGD on noisy MNIST · MNIST | pristine-example learning rate | Hurt |
| Increase per-coordinate gradient winsorization (ablation) | Winsorized SGD · MNIST with label noise | overfit growth rate | Helped |
| Use winsorization with c > 1 (ablation) | Winsorized SGD · MNIST with label noise | training accuracy relative to proper accuracy | Hurt |
| Use large amounts of gradient winsorization (ablation) | Winsorized SGD · MNIST with label noise | training and test accuracy | Mixed |
Closest papers by meaning
- Gradient Starvation: A Learning Proclivity in Neural Networks (NeurIPS 2021 · similarity 0.90)
- Learning Gradient Descent: Better Generalization and Longer Horizons (ICML 2017 · similarity 0.90)
- No Wrong Turns: The Simple Geometry Of Neural Networks Optimization Paths (ICML 2024 · similarity 0.89)
- Deep Linear Network Training Dynamics from Random Initialization: Data, Width, Depth, and Hyperparameter Transfer (ICML 2025 · similarity 0.89)
- Do Neural Networks Need Gradient Descent to Generalize? A Theoretical Study (NeurIPS 2025 · similarity 0.89)
- Bias of Stochastic Gradient Descent or the Architecture: Disentangling the Effects of Overparameterization of Neural Networks (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.