arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2610.20187cs.LGmath.OC
收藏

指数凹在线学习中的维度最优动态遗憾

Dimension-Optimal Dynamic Regret in Exp-Concave Online Learning

Peking University(北京大学)

Yanyi Li, Wendao Wu, Haihan Zhang, Qinshi Wang, Chenheng Zhang, Chunyuan Zheng, Cong Fang, Haoxuan Li, Zhouchen Lin

AI总结:

该研究解决指数凹在线学习中维度最优动态遗憾问题,提出含d^{1/3}弯曲维度因子的学习器,获匹配上下界,改进现有维度因子,提升了在线学习的理论边界。

摘要:

追踪变化的比较器是在线学习中的核心问题。对于指数凹损失,此前研究已确定固定维度下遗憾对时间范围T和欧氏路径长度P_T的依赖关系为T^{1/3}P_T^{2/3},但维度的最优代价仍未解决。我们通过建立尖锐的d^{1/3}弯曲维度因子解决了该问题,改进了近期切换和区间约简工作[1,2]中的d^{2/3}因子。为实现该改进,我们在R^d的紧凸子集上开发了一种合适的全信息学习器,该学习器可适应未知路径长度,且在固定的与维度无关的直径、梯度和曲率参数下,保证Reg_T(u_{1:T}) ≤ Õ(d + min{d^{1/3}T^{1/3}P_T^{2/3}, √(TP_T), T})。在直径为1的欧氏球上的匹配下界确定了极小极大前沿(对数因子除外),当T ≳ d^2时,包括从弯曲速率到更大路径预算的转变。该下界已适用于每轮仅依赖一个坐标的损失,分离了跨方向学习的不可避免代价。我们的上界源于比较器运动的平方根高斯KL控制和补偿历史重启的抵消论证。弯曲学习器维护O(log²T)个高斯状态,每轮使用一个梯度,且在马氏距离投影神谕下运行时间接近线性于T。

英文摘要:

Tracking changing comparators is a central problem in online learning. For exp-concave losses, previous work established the $T^{1/3}P_T^{2/3}$ dependence on the horizon $T$ and Euclidean path length $P_T$ for fixed dimension, while the optimal cost of dimension remained unresolved. We resolve this question by establishing a sharp $d^{1/3}$ curved dimension factor, improving the $d^{2/3}$ factor of recent switching and interval reductions [1, 2]. To achieve this improvement, we develop a proper, full-information learner on compact convex subsets of $\mathbb{R}^d$ that adapts to the unknown path length and, under fixed dimension-independent diameter, gradient, and curvature parameters, guarantees $$\operatorname{Reg}_T(u_{1:T}) \le \widetilde{O}\left(d+\min\left\{d^{1/3}T^{1/3}P_T^{2/3},\sqrt{TP_T},T\right\}\right).$$ A matching lower bound on a diameter-one Euclidean ball establishes the minimax frontier up to logarithmic factors when $T\gtrsim d^2$, including the transition from the curved rate to larger path budgets. The lower bound already holds for losses that depend on only one coordinate per round, isolating an unavoidable cost of learning across directions. Our upper bound follows from square-root Gaussian KL control of comparator movement and a cancellation argument that pays for history restarts. The curved learner maintains $O(\log^2 T)$ Gaussian states, uses one gradient per round, and runs in time near-linear in $T$ under a Mahalanobis-projection oracle.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑