The relationship between randomness and power-law distributed move lengths in random walk algorithms
The relationship between randomness and power-law distributed move lengths in random walk algorithms
复制标题
随机游走算法中随机性与幂律分布移动长度之间的关系
DOI：
10.1016/j.physa.2014.01.060
复制
发表时间：
2014
期刊：
影响因子：
3.3
通讯作者：
Yukio-Pegio Gunji
中科院分区：
文献类型：
--
作者：
Tomoko Sakiyama;Yukio-Pegio Gunji
最近，我们提出了一种新的随机行走算法，称为REV算法，其中代理使用最近的四个随机数来改变控制它的方向规则。在这里，我们研究了一个非有界数，即关于移动方向的“随机性”，对于最优搜索和规则变化中的幂分布步长是如何重要的。我们提出了两种算法：REV算法和REV有界算法。在REV算法中，用于更改规则的四个随机数中的一个是无界的。相反，Rev-Bound算法中的所有四个随机数都是有界的。结果表明，REV算法具有更一致的幂分布步长和更灵活的搜索行为。
Recently, we proposed a new random walk algorithm, termed the REV algorithm, in which the agent alters the directional rule that governs it using the most recent four random numbers. Here, we examined how a non-bounded number, i.e., “randomness” regarding move direction, was important for optimal searching and power-law distributed step lengths in rule change. We proposed two algorithms: the REV and REV-bounded algorithms. In the REV algorithm, one of the four random numbers used to change the rule is non-bounded. In contrast, all four random numbers in the REV-bounded algorithm are bounded. We showed that the REV algorithm exhibited more consistent power-law distributed step lengths and flexible searching behavior.