神经网络的非零温蒙特卡洛及迂回的蒙特卡洛学习算法

A NON-ZERO TEMPERATURE MONTE CARLO ALGORITHM AND A OUTFLANKING MONTE CARL ALGORITHM FOR NEURAL NETWORKS

本文在人工神经网络的学习中选用适当的评价函数实现了将零温度的蒙特卡洛算法［２］向非零温情况的推广，得到优于零温的结果，但是仍未克服其学习结果依赖于初值的缺点．在分析其能量特点的基础上，本文提供了非零温的蒙特卡洛迂回学习算法，它能够成功地摆脱初值的影响，获得高存储容量的神经网络，此结论已经为仿真研究的结果所证实．

A suitable appraisal function have been proposed and a working to spread zero temperature Monte Carlo algorithm over non-zero temperature Monte Carlo algorithm havebeen accomplished,the result of the non-zero temperature Monte Carlo algorithm is superior to zero temperature,but,the result also depends on the initial value of the interconnect weight.Based on analyseing characteristic for the energy,a outflanking Monte Carlo algorithm have been proposed,the result of that is independent of the initial value of the interconnect weight and may obtain a neural network having high storage capacity.The above have been verifyed by computer simulation.