Primal Acceleration of Newton's Method
A Newton-style optimizer claims an `O(1/k^3)` global rate with only one linear solve per step.
Nikita Doikov’s paper gives a direct accelerated Newton method for convex functions with Lipschitz-continuous Hessians. The method stays in primal variables and avoids auxiliary nonlinear regularized subproblems, parameter searches, and dual extragradient corrections. The abstract says it can also be run Hessian-free with an inexact linear solver while preserving the stated rate. Extensions cover Bregman geometry and composite optimization problems. ArXiv · AI/CL/LG's note
Nikita Doikov’s paper gives a direct accelerated Newton method for convex functions with Lipschitz-continuous Hessians. The method stays in primal variables and avoids auxiliary nonlinear regularized subproblems, parameter searches, and dual extragradient corrections. The abstract says it can also be run Hessian-free with an inexact linear solver while preserving the stated rate. Extensions cover Bregman geometry and composite optimization problems. ArXiv · AI/CL/LG's note
score 5