Overfitting of Spectral Gradient Descent: How Matrix Geometry shapes Generalization and Implicit Bias
arXiv:2609.32270v1 Announce Type: new Abstract: We study the generalization of spectral gradient descent (SpecGD) in overparameterized matrix classification with corrupted labels. Each input combines a shared low-rank signal with a rank-one sample-specific perturbation, referred to as a shortcut, t…
Read the full story at arXiv stat.ML ↗
Timeline · 1 report
- 2026-09-29 04:00 · arXiv stat.ML
Overfitting of Spectral Gradient Descent: How Matrix Geometry shapes Generalization and Implicit Bias