重要性采样(英语:importance sampling)是统计学中估计某一分布性质时使用的一种方法。该方法从与原分布不同的另一个分布中采样,而对原先分布的性质进行估计。重要性采样与计算物理学中的伞形采样相关。
原理
假设为概率空间上的一个随机变量。我们想要估计X的期望值,记作E[X;P]。如果根据P随机抽取样本,估计的期望值即
这一估计的精确度取决于X的方差,
而重要性采样的基本思想则是从另一个分布中抽取样本,用以降低E[X;P]估计的方差。进行重要性采样时,首先选择一个随机变量,使得E[L;P]=1,并满足P上几乎处处。由此,可以定义新的概率
于是,我们可以从P(L)上抽样,通过变量X/L估计E[X;P]。如果成立,此时的估计便优于直接在原分布上采样得到的估计。
当X在Ω上不变号时,最优的L为。此时X/L*即为要估计的E[X;P],只需一个样本便可得到该值。然而由于L*与要估计的E[X;P]有关,在实际操作中我们无法取到理论上最优的L*。不过,我们仍可以采用如下方式逼近该理论值:
于是,要估计的期望值可改写为:
注意到,更优(即让估计值方差更小)的P(L)会使得样本分布的频率与其在E[X;P]计算中的权重更加相关。这也是该方法得名“重要性采样”的原因。
参考文献
- Arouna, Bouhari. Adaptative Monte Carlo Method, A Variance Reduction Technique. Monte Carlo Methods and Their Applications. 2004, 10 (1): 1–24. doi:10.1515/156939604323091180.
- Bucklew, James Antonio. Introduction to Rare Event Simulation. New York: Springer-Verlag. 2004.
- Doucet, A.; de Freitas, N.; Gordon, N. Sequential Monte Carlo Methods in Practice. Springer. 2001. ISBN 978-0-387-95146-1.
- Ferrari, M.; Bellini, S. Importance Sampling simulation of turbo product codes. The IEEE International Conference on Communications. 2001, 9: 2773–2777. doi:10.1109/ICC.2001.936655.
- Mazonka, Oleg. Easy as Pi: The Importance Sampling Method (PDF). Journal of Reference. 2016, 16 [2017-09-18]. (原始内容存档 (PDF)于2021-04-14).
- Oberg, Tommy. Modulation, Detection, and Coding. New York: John Wiley & Sons. 2001.
- Press, WH; Teukolsky, SA; Vetterling, WT; Flannery, BP. Section 7.9.1 Importance Sampling. Numerical Recipes: The Art of Scientific Computing 3rd. New York: Cambridge University Press. 2007 [2017-09-18]. ISBN 978-0-521-88068-8. (原始内容存档于2011-08-11).
- Ripley, B. D. Stochastic Simulation. Wiley & Sons. 1987.
- Smith, P. J.; Shafi, M.; Gao, H. Quick simulation: A review of importance sampling techniques in communication systems. IEEE J. Select. Areas Commun. 1997, 15 (4): 597–613. doi:10.1109/49.585771.
- Srinivasan, R. Importance sampling – Applications in communications and detection. Berlin: Springer-Verlag. 2002.
Wikiwand in your browser!
Seamless Wikipedia browsing. On steroids.
Every time you click a link to Wikipedia, Wiktionary or Wikiquote in your browser's search results, it will show the modern Wikiwand interface.
Wikiwand extension is a five stars, simple, with minimum permission required to keep your browsing private, safe and transparent.