论文部分内容阅读
数据互联是确定多目标跟踪算法中的量测源和分配概率β_i~t问题,其中β_i~t是表示第i个量测来自第t个目标这一事件的概率。本文介绍使用一个分层博尔兹曼机求解数据互联问题的新并行计算方法。研究证明,如果能得到充分多的二元神经元层数,就可以用任意小的误差计算互联概率。特别是,证明了概率β_i~j等于分层的两维网络中神经元v(i,j)激发的相对频率。文中还介绍一些简单的跟踪例子,这些例子将对用于精确的数据互联解的博尔兹曼算法的性能与使用Hopfield神经网络的另一种并行算法的性能进行比较。
The data interconnection is to determine the measurement source and the distribution probability β_i ~ t in the multi-target tracking algorithm, where β_i ~ t is the probability of the i-th measurement from the t-th target. This article describes a new parallel method of computing data interconnection using a hierarchical Boltzmann machine. Research shows that if we can get enough layers of binary neurons, we can use any small error to calculate the probability of interconnection. In particular, it is proved that the probability β_i ~ j is equal to the relative frequency excited by the neuron v (i, j) in a two-dimensional hierarchical network. The article also introduces some simple tracking examples that compare the performance of the Boltzmann algorithm used for accurate data interconnection solutions to the performance of another parallel algorithm using Hopfield neural networks.