Regression-Based Single-Point Zeroth-Order Optimization (2507.04223v1)

Abstract: Zeroth-order optimization (ZO) is widely used for solving black-box optimization and control problems. In particular, single-point ZO (SZO) is well-suited to online or dynamic problem settings due to its requirement of only a single function evaluation per iteration. However, SZO suffers from high gradient estimation variance and slow convergence, which severely limit its practical applicability. Despite recent advances, the convergence of existing SZO methods remains inferior to that of two-point ZO methods. To overcome this limitation, we propose a novel yet simple SZO framework, termed regression-based SZO (RESZO), which substantially enhances the convergence rate. Unlike conventional ZO methods that rely solely on function evaluations at the current point for gradient estimation, RESZO improves gradient estimation by effectively leveraging historical function evaluations from previous iterations. Specifically, RESZO constructs a surrogate function via regression using recent historical evaluations and employs the gradient of this surrogate function for iterative updates. Two variants of RESZO, which fit linear and quadratic surrogate functions respectively, are introduced. Theoretically, we provide a non-asymptotic convergence analysis for the linear variant of RESZO, showing that its convergence rates are comparable to those of two-point ZO methods for both smooth nonconvex and strongly convex functions. Moreover, extensive numerical experiments demonstrate that RESZO not only matches but empirically outperforms two-point ZO in terms of function query complexity.