SOURCE-LINKED INTELLIGENCE

Inference for Newton Methods with Accelerated Sketch-and-Project via Random Scaling

arXiv · AI, language, vision and robotics · article · Sep 11, 2026 · UTC

We study an online sketched Newton method that approximates the Newton direction at each step via a state-of-the-art sketching solver, called the generalized accelerated sketch-and-project solver (GAS), thereby mitigating the computational bottleneck of classical second-order methods. The GAS solver improves upon vanilla, unaccelerated sketch-and-project solvers by achieving accelerated convergence through Nesterov momentum updates, and accommodates a flexible projection metric whose proper choice further reduces computational cost. Building on this design, we establish asymptotic normality of

Read original source ↗ Open in workspace

recordType

paper

region

Global