A Newton-MR algorithm with complexity guarantees for nonconvex smooth unconstrained optimization

Code

A brief intro: We propose a Newton-MR variant that can enjoy both the global optimal complexity result and the local quadratic convergence rate.