Study Shows Gradient Averaging Reaches Optimal Statistical Efficiency in Newton Sketch Methods
A new paper analyzes statistical inference for online sketched Newton methods, where each Newton system is approximately solved via inner randomized sketching, and shows gradient averaging achieves optimal statistical efficiency.
Read the full story at arXiv stat.ML ↗
Timeline · 1 report
- 2026-11-19 05:00 · arXiv stat.ML
A Free Lunch: Gradient Averaging Achieves Optimal Statistical Efficiency in Newton Sketch