A multi-Markovian switching-based strategy for solving the stochastic point location problem
A multi-Markovian switching-based strategy for solving the stochastic point location problem
复制标题
DOI：
10.1007/s00521-022-06894-2
复制
发表时间：
2022-03
影响因子：
6
通讯作者：
Ying Guo;Shenghong Li
中科院分区：
文献类型：
--
作者：
Ying Guo;Shenghong Li
随机点定位（Stochastic Point Location）问题是机器学习中广泛存在的一个问题，它考虑到学习实体（即机制、算法等）试图通过与随机环境的交互来定位某个点。常规技术是将搜索空间采样为离散点并执行随机游走。然而，随机游走仅限于相邻点。本文提出了一种基于随机游走的三层算法的扩展版本，以克服上述缺陷。具体来说，该算法利用多马尔可夫开关推广的随机游走相邻节点的间歇节点。因此，整个方法可以被看作是马尔可夫链，并可以构造其转换矩阵，然后通过严格的数学过程的收敛性。实验结果表明，该算法具有较强的稳定性、较高的精度和较快的计算速度。
Stochastic Point Location problem considering that a learning entity (i.e. mechanisms, algorithm, etc) attempts to locate a certain point by interaction with a stochastic environment is encountered widely in Machine Learning. A conventional technique is to sample the search space into discrete points and perform a random walk. Nevertheless, the random walk is confined to the neighboring point. In this paper, an extended version of the random walk-based triple level algorithm is introduced to overcome the aforementioned defect. Specifically, the proposed algorithm exploits the multi-Markovian switching to generalize the random walk concerning adjacent nodes to intermittent nodes. Hence, the whole approach could be regarded as the Markov chain, and its transform matrix could be constructed, followed by a rigorous mathematical pf procedure of the convergence. The experimental results demonstrate the effectiveness and efficiency of the proposed algorithm, showing its abilities of stronger stability, a higher precision, and a faster speed in comparison with the counterparts available in open literatures.