通过左到右最小值指数倾斜分布研究非均匀到达的秘书问题
The secretary problem with non-uniform arrivals via a left-to-right-minimum exponentially tilted distribution
AI总结:
本文研究左到右最小值指数倾斜分布下的非均匀到达秘书问题,刻画所有 $q_n$ 区间中最优拒绝比例与极限成功概率,并给出成功概率等于或大于 $e^{-1}$ 的条件。
摘要:
我们解决了排序物品以统计有偏顺序而非均匀随机顺序到达情形下的秘书问题。该偏差由参数为 $q\in(0,\infty)$ 的左到右最小值指数倾斜分布给出。即,对于 $σ\in S_n$,$P_n(σ)$ 正比于 $q^{\text{LR}^{-}_n(σ)}$,其中左到右最小值统计量 $\text{LR}^-_n$ 定义为 $$\text{LR}^{-}_n(σ)=|\{j\in[n]: σ_j=\min\{σ_i:1\le i\le j\}\}|,\\ σ\in S_n.$$ 当 $q\in(0,1)$ 时,排名更高的物品相比均匀分布情形倾向于更早到达;而当 $q\in(1,\infty)$ 时,它们倾向于更晚到达。在经典问题中,渐近最优策略是拒绝前 $M_n^*$ 个物品,其中 $M_n^*\sim\frac ne$,然后选择排名高于前 $M_n^*$ 个物品中任何一个的首个物品(如果存在这样的物品)。这产生 $e^{-1}$ 的极限成功概率。在上述到达偏差下,我们针对 $\{q_n\}_{n=1}^\infty$ 的所有区间,计算了最优策略 $M_n^*$ 的渐近行为及相应的极限成功概率。特别地,如果 $\{q_n\}_{n=1}^\infty$ 的主导阶渐近行为至少为 $\frac1{\log n}$,并且其阶不超过 $o(n)$,则使用渐近最优策略时的极限成功概率为 $e^{-1}$;否则,该极限成功概率大于 $e^{-1}$。此外,被渐近最优策略概括性拒绝的数量的极限比例 $\lim_{n\to\infty}\frac{M^*_n}n$ 位于 $(0,1)$ 中,当且仅当 $\lim_{n\to\infty}q_n\in(0,\infty)$。
英文摘要:
We solve the secretary problem in the case that the ranked items arrive in a statistically biased order rather than in uniformly random order. The bias is given by the left-to-right-minimum exponentially tilted distribution with parameter $q\in(0,\infty)$. That is, for $σ\in S_n$, $P_n(σ)$ is proportional to $q^{\text{LR}^{-}_n(σ)}$, where the left-to-right minimum statistic $\text{LR}^-_n$ is defined by $$ \text{LR}^{-}_n(σ)=|\{j\in[n]: σ_j=\min\{σ_i:1\le i\le j\}\}|,\ σ\in S_n. $$ For $q\in(0,1)$, higher ranked items tend to arrive earlier than in the case of the uniform distribution, and for $q\in(1,\infty)$, they tend to arrive later. In the classical problem, the asymptotically optimal strategy is to reject the first $M_n^*$ items, where $M_n^*\sim\frac ne$, and then to select the first item ranked higher than any of the first $M_n^*$ items (if such an item exists). This yields $e^{-1}$ as the limiting probability of success. With the above bias on arrivals, we calculate the asymptotic behavior of the optimal strategy $M_n^*$ and the corresponding limiting probability of success, for all regimes of $\{q_n\}_{n=1}^\infty$. In particular, if the leading order asymptotic behavior of $\{q_n\}_{n=1}^\infty$ is at least $\frac1{\log n}$, and if also its order is no more than $o(n)$, then the limiting probability of success when using an asymptotically optimal strategy is $e^{-1}$; otherwise, this limiting probability of success is greater than $e^{-1}$. Also, the limiting fraction of numbers, $\lim_{n\to\infty}\frac{M^*_n}n$, that are summarily rejected by an asymptotically optimal strategy lies in $(0,1)$ if and only if $\lim_{n\to\infty}q_n\in(0,\infty)$.
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。