Research map / Optimization / Implicit bias and generalization
On the Implicit Bias of Initialization Shape: Beyond Infinitesimal Mirror Descent
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 the initialization shape toward an imbalanced ratio between layers | Two-layer diagonal linear networks · Sparse regression | implicit-bias regime | Mixed |
| Use balanced initialization rather than imbalanced initialization | Two-layer diagonal and fully connected linear networks · Sparse regression | population error | Helped |
| Decrease initialization scale while holding shape fixed | Two-layer diagonal linear networks · Square-loss regression | implicit-bias regime | Mixed |
| Take vanishing initialization | Fully connected linear networks with multiple neurons · Square-loss regression | implicit regularizer | Mixed |
| Use the warped infinitesimal mirror descent construction instead of requiring the dynamics metric to be a Hessian map | Implicit-bias analysis for fully connected linear networks · Square-loss regression | implicit-bias characterization | Helped |
Closest papers by meaning
- Mirror, Mirror of the Flow: How Does Regularization Shape Implicit Bias? (ICML 2025 · similarity 0.89)
- Implicit Bias of Mirror Flow for Shallow Neural Networks in Univariate Regression (ICLR 2025 · similarity 0.89)
- Mask in the Mirror: Implicit Sparsification (ICLR 2025 · similarity 0.88)
- From Lazy to Rich: Exact Learning Dynamics in Deep Linear Networks (ICLR 2025 · similarity 0.87)
- Implicit Regularization for Tubal Tensor Factorizations via Gradient Descent (ICML 2025 · similarity 0.87)
- VI^3NR: Variance Informed Initialization for Implicit Neural Representations (CVPR 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.