论文部分内容阅读
在多目标跟踪情况下求解数据互联问题要求计算把第i个量测分配给第j个目标的概率β~j。先前,我们曾提出一个根据分层的、异步(序列的)博尔兹曼机的并行结构估计互联概率。本文介绍具有随机神经元的这种结构的有效模拟实现。用矢量郎之万方程描述这种网络的动态,结果网络近似为一个真正的具有潜在快速收敛性的同步博尔兹曼机。在分层的两维网络中,概率β~ji渐近地等于量化的神经元输出vij的激活频率。描述了近似真正互联概率的设计准则。分析了在有界区域内表示为扩散过程的每一个随机神经元的暂态和稳态性能。把分层扩散网络的性能与理论的极限作了比较,也同异步博尔兹曼机的性能进行了比较。
Solving the data interconnection problem in the case of multi-target tracking requires calculating the probability β ~ j that the ith measurement is assigned to the jth target. Previously, we proposed an estimate of the interconnection probability based on the parallel structure of a hierarchical, asynchronous (sequence) Boltzmann machine. This article describes an effective simulation of this structure with random neurons. The vector Lang’s equation is used to describe the dynamics of this network. The resulting network approximates a true synchronized Boltzmann machine with potentially fast convergence. In a layered two-dimensional network, the probability β ~ ji is asymptotically equal to the activation frequency of the quantized neuron output vij. Describes the design criteria for approximate real interconnectivity. The transient and steady state properties of each random neuron expressed as a diffusion process in a bounded region are analyzed. The performance of stratified diffusion networks is compared with the theoretical limit, and it is also compared with the performance of asynchronous Bowlzman machines.