跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2211.03997 · v4

Online Nonconvex Decision Making with Approximate Oracles

首次提交 2022-11-08 12:28
版本更新 2026-11-13 10:53
雷达观测 2026-11-17 04:57
arXiv ↗PDF ↗

摘要

We study online decision making with possibly nonconvex (for example, combinatorial or integer) local feasible sets and convex long-term constraints. At each period, a decision maker selects a local action to maximize cumulative reward while ensuring that the sequence of actions satisfies the global constraints in the long term. It is known that O(T)O(\sqrt{T}) long-term constraint violation and O(T)O(\sqrt{T}) reward regret are achievable if given access to a local optimization oracle that exactly solves optimization problems over each local feasible set. Instead of requiring exact optimization over each local feasible set, we assume access to an approximate local optimization oracle that permits optimization errors quantified by two parameters (δ+,δ×)(δ_+,δ_\times). We develop an approximate online dual mirror descent algorithm and show that its guarantees depend on a feasibility slack parameter d‾\underline d, which measures the compatibility between local feasible sets and the long-term constraint. Under the standard random permutation model, when δ+=O(1)δ_+=O(1), δ×<d‾δ_\times<\underline d and d‾−δ×=Ω(1)\underline d-δ_\times=Ω(1), the algorithm achieves O(T)O(\sqrt T) long-term constraint violation and reward regret up to O(T+δ×T)+δ+TO\bigl(\sqrt T+δ_\times T\bigr)+δ_+ T. We also show a matching limitation: when δ+=0δ_+=0 and δ×>d‾δ_\times>\underline d, no online algorithm relying on such approximate oracle outputs can guarantee sublinear long-term constraint violation in general.

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

本站首次收录为 v4,暂无更早版本可对照。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v4 · 2026-11-13 10:53 · 最新观测版本

Online Nonconvex Decision Making with Approximate Oracles

We study online decision making with possibly nonconvex (for example, combinatorial or integer) local feasible sets and convex long-term constraints. At each period, a decision maker selects a local action to maximize cumulative reward while ensuring that the sequence of actions satisfies the global constraints in the long term. It is known that O(T)O(\sqrt{T}) long-term constraint violation and O(T)O(\sqrt{T}) reward regret are achievable if given access to a local optimization oracle that exactly solves optimization problems over each local feasible set. Instead of requiring exact optimization over each local feasible set, we assume access to an approximate local optimization oracle that permits optimization errors quantified by two parameters (δ+,δ×)(δ_+,δ_\times). We develop an approximate online dual mirror descent algorithm and show that its guarantees depend on a feasibility slack parameter d‾\underline d, which measures the compatibility between local feasible sets and the long-term constraint. Under the standard random permutation model, when δ+=O(1)δ_+=O(1), δ×<d‾δ_\times<\underline d and d‾−δ×=Ω(1)\underline d-δ_\times=Ω(1), the algorithm achieves O(T)O(\sqrt T) long-term constraint violation and reward regret up to O(T+δ×T)+δ+TO\bigl(\sqrt T+δ_\times T\bigr)+δ_+ T. We also show a matching limitation: when δ+=0δ_+=0 and δ×>d‾δ_\times>\underline d, no online algorithm relying on such approximate oracle outputs can guarantee sublinear long-term constraint violation in general.

查看此版本解读 →查看此版本原文 ↗