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 2303.01453cs.DScs.LG
收藏

改进的专家学习空间边界

Improved Space Bounds for Learning with Experts

MIT(麻省理工学院);Northeastern University(东北大学)

Anders Aamand, Justin Y. Chen, Huy Lê Nguyen, Sandeep Silwal

摘要:

我们改进了空间和遗憾之间的权衡,为在线学习与专家意见的问题超过$T$天与$n$专家。在给定$δ\in (0,1)$的$n^δ$空间预算的情况下,我们给出了一个实现后悔$\tilde{O}(n^2 T^{1/(1+δ)})$的算法,改进了[PZ23]中的后悔界$\tilde{O}(n^2 T^{2/(2+δ)})$.改进是特别突出的政权$δ\rightarrow 1$的遗憾,我们的算法接近$\tilde{O}_n(\sqrt{T})$,匹配的$T$依赖在标准的在线设置没有空间限制。

英文摘要:

We give improved tradeoffs between space and regret for the online learning with expert advice problem over $T$ days with $n$ experts. Given a space budget of $n^δ$ for $δ\in (0,1)$, we provide an algorithm achieving regret $\tilde{O}(n^2 T^{1/(1+δ)})$, improving upon the regret bound $\tilde{O}(n^2 T^{2/(2+δ)})$ in the recent work of [PZ23]. The improvement is particularly salient in the regime $δ\rightarrow 1$ where the regret of our algorithm approaches $\tilde{O}_n(\sqrt{T})$, matching the $T$ dependence in the standard online setting without space restrictions.

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。

↑