Thanh Vo Hong
Department of Computer Science, Aalto University, Espoo, Finland and The Microsoft Research - University of Trento Centre for Computational and Systems Biology (COSBI) Rovereto, Italy.
IET Syst Biol. 2019 Feb;13(1):16-23. doi: 10.1049/iet-syb.2018.5035.
We investigate the computational challenge of improving the accuracy of the stochastic simulation estimation by inducing negative correlation through the anticorrelated variance reduction technique. A direct application of the technique to the stochastic simulation algorithm (SSA), employing the inverse transformation, is not efficient for simulating large networks because its computational cost is similar to the sum of independent simulation runs. We propose in this study a new algorithm that employs the propensity bounds of reactions, introduced recently in their rejection-based SSA, to correlate and synchronise the trajectories during the simulation. The selection of reaction firings by our approach is exact due to the rejection-based mechanism. In addition, by applying the anticorrelated variance technique to select reaction firings, our approach can induce substantial correlation between realisations, hence reducing the variance of the estimator. The computational advantage of our rejection-based approach in comparison with the traditional inverse transformation is that it only needs to maintain a single data structure storing propensity bounds of reactions, which is updated infrequently, hence achieving better performance.
我们研究了通过反相关方差缩减技术引入负相关来提高随机模拟估计准确性的计算挑战。将该技术直接应用于采用逆变换的随机模拟算法（SSA），对于模拟大型网络效率不高，因为其计算成本与独立模拟运行的总和相似。在本研究中，我们提出了一种新算法，该算法利用最近在基于拒绝的SSA中引入的反应倾向边界，在模拟过程中关联并同步轨迹。由于基于拒绝的机制，我们的方法对反应触发的选择是精确的。此外，通过应用反相关方差技术来选择反应触发，我们的方法可以在实现之间诱导显著的相关性，从而降低估计器的方差。与传统逆变换相比，我们基于拒绝的方法的计算优势在于它只需要维护一个存储反应倾向边界的单一数据结构，该结构很少更新，因此具有更好的性能。
