Updated Paper Examines How Correlations Undermine Gradient Descent
A revised arXiv paper argues that correlations in data degrade gradient descent, building on natural gradient descent and preconditioned optimizers that estimate curvature to correct it.
Read the full story at arXiv cs.LG ↗
Timeline · 1 report
- 2026-11-04 05:00 · arXiv cs.LG
Correlations Are Ruining Your Gradient Descent