Research map / Optimization / Implicit bias and generalization
On the Effect of Initialization: The Scaling Path of 2-Layer Neural Networks
Sebastian Neumayer, Lénaïc Chizat, Michael Unser
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| match the gradient-descent initialization scale beta to the scaling-path scale alpha (ablation) | gradient descent training · 2D interpolation, 10 samples | regularized W2 objective gap: beta=alpha gives minimum for large alpha, e.g. 5.66e2 at alpha=3.3e0 vs 1.79e3 at beta=3.3e0, alpha=1e1 | Improved |
| use gradient descent solution instead of the variational scaling-path solution | 2-layer ReLU interpolation · 2D interpolation, 10 samples | functional value: 7.50e1 -> 9.36e1 at alpha=1e-1 | Hurt |
| increase regularization scale alpha | scaling path · 2D interpolation, 10 samples | smoothness of solution | Improved |
| increase alpha toward infinity | scaling path · 2D interpolation, 10 samples | convergence to kernel optimum: 7.73e2 -> 2.79e2 vs 2.83e2 NTK value | Improved |
| decrease alpha and beta toward small values | gradient descent vs scaling path · 2D interpolation, 10 samples | similarity of solutions | Hurt |
| restrict support of measure to the smaller grid P instead of Q (ablation) | scaling path discretization · 2D interpolation, 10 samples | solution quality | No clear effect |
Closest papers by meaning
- Training Two-Layer ReLU Networks with Gradient Descent is Inconsistent (JMLR 2022 · similarity 0.89)
- The Hidden Convex Optimization Landscape of Regularized Two-Layer ReLU Networks: an Exact Characterization of Optimal Solutions (ICLR 2022 · similarity 0.88)
- Gradient Descent Provably Optimizes Over-parameterized Neural Networks (ICLR 2019 · similarity 0.88)
- Large Stepsize Gradient Descent for Non-Homogeneous Two-Layer Networks: Margin Improvement and Fast Optimization (NeurIPS 2024 · similarity 0.88)
- Simplicity Bias and Optimization Threshold in Two-Layer ReLU Networks (ICML 2025 · similarity 0.88)
- Convex Formulations for Training Two-Layer ReLU Neural Networks (ICLR 2025 · similarity 0.87)
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.
