一种新型人工神经网络的实现与改进——概率逻辑神经网络  

Implementation and Improvement of a Noval Artificial Neural Network——Probabilistic Logic Neural Network

作　　者：李春燕[1] 陈松灿[1] 杨国庆[1] 

机构地区：[1]南京航空学院计算机科学与工程系

出　　处：《南京航空航天大学学报》1991年第S1期26-32,共7页Journal of Nanjing University of Aeronautics & Astronautics

基　　金：航空基金的资助

摘　　要：概率逻辑神经网络模型是一新兴的神经网络模型,它不仅具有较快的训练速度,而且具有“自退火”性能,并易于硬件实现,从而成功地克服了误差反向传播(BP)模型的训练时间长,易于陷于局部最小点的弱点。本文将给出概率逻辑神经网络(PLNN)的一般结构、学习算法及其模拟实现的情况。并由此推广到多层联想网络的实现与分析,最后给出了PLNN与BP的比较结果。The model of probabilistic logic neural network (PLNN) is a novel one. It not only has higher training speed and self-annealing property,but also can be easily realized with hardware. Hence.it can efficiently overcome the weakness of the error back-propagation (BP) network which takes a long time to train and falls into the local minimum easily. The paper will introduce the general structure,learning algorithm and simulation of PLNN and discuss the accomplishment and the analysis of multilayered associative network. Finally, the brief performance comparison results between PLNN and BP are given.

关 键 词：概率逻辑 神经网络 相联处理 BP模型 神经计算机 

分 类 号：TP399[自动化与计算机技术—计算机应用技术] O211.62[自动化与计算机技术—计算机科学与技术]