On speeding up stochastic simulations by parallelization of random number generation.
On speeding up stochastic simulations by parallelization of random number generation.
复制标题
通过随机数生成的并行化来加速随机模拟。
DOI：
10.1016/j.ces.2015.06.066
复制
发表时间：
2015
影响因子：
4.7
通讯作者：
Ramkrishna,Doraiswami
中科院分区：
文献类型：
--
作者：
Shu,Che-Chi;Tran,Vu;Binagia,Jeremy;Ramkrishna,Doraiswami
本文基于一个非常简单的想法添加到随机模拟的工具包中。它适用于SSA和Tau-leap算法，可以显着减少计算时间。随机模拟基于计算样本路径，该样本路径基于随机数的生成，该随机数具有如SSA（吉莱斯皮，1977）中精确规定的分布函数或静止间隔方法（Shah等人，1977）或分布函数，其特征在于设计用于提高效率的近似（如在Tau-leap算法中（Cao等人，2006，Tian和Burrage，2004，Peng等人，2007，吉莱斯皮，2001，Ramkrishna等人，2014），其中使用了具有参数R1的跳跃条件）。通常的策略涉及顺序计算大量的样本路径在一个有界的时间间隔，这是由一组离散的时间子区间随机数生成所覆盖。这里的策略通过并行化样本路径集合的随机子间隔的生成，直到已经针对所述时间间隔计算了所有样本路径，而与前述不同。该过程的优点在于，随机数发生器的启动时间显著减少。许多例子表明，从SSA以及Tau-飞跃算法建立的方法的优点是远远超过概念。
This paper adds to the tool kit of stochastic simulations based on a very simple idea. Applicable to both SSA and Tau-leap algorithms, it can notably reduce computational times. Stochastic simulations are based on computing sample paths based on the generation of random numbers with either exactly stipulated distribution functions as in SSA (Gillespie, 1977) or in the method of interval of quiescence (Shah et al., 1977) or distribution functions featuring approximations designed to promote efficiency (as in Tau-leap algorithms (Cao et al., 2006, Tian and Burrage, 2004, Peng et al., 2007, Gillespie, 2001, Ramkrishna et al., 2014) where a leap condition with the parameter epsilon is used). The usual strategy involves sequential computation of a large number of sample paths over a bounded time interval which is covered by a set of discrete time subintervals obtained by random number generation. The strategy here departs from the foregoing by parallelizing the generation of random subintervals for the set of sample paths until all sample paths have been computed for the stated time interval. The advantage of this procedure lies in the fact that the time for initiation of the random number generator has been notably reduced. Many examples are demonstrated from SSA as well as Tau-leap algorithms to establish that the advantage of the approach is much more than conceptual.