Abstract:Many nonsmooth, nonconvex objectives in learning and signal recovery are $\rho$-weakly convex. We minimize such a function in deterministic and stochastic settings when the weak-convexity parameter $\rho$ is unknown. The objective is not required to be globally Lipschitz continuous or smooth. We propose the Adaptive Prox-Guided Scheme (APS), a one-trial proximal algorithm that adapts the proximal parameter online and bidirectionally through a descent test, allowing it to exploit favorable local structure.
In the deterministic setting, APS obtains an $O(\varepsilon^{-2})$ iteration complexity for producing an $\varepsilon$-subgradient stationary point. In the stochastic setting, APS achieves a high-probability $O(\varepsilon^{-2})$ iteration bound for driving the Moreau-envelope gradient below $\varepsilon$. This result holds under deliberately weak oracle assumptions: the function-difference estimates may be biased and heavy-tailed, and the stochastic proximal oracle need only be sufficiently accurate with constant probability when the proximal parameter lies below $1/(2\rho)$ (unknown to the algorithm), and can be arbitrary otherwise.
From: Miaolan Xie [view email]
[v1]
Mon, 15 Jun 2026 20:49:56 UTC (30 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。