【摘要】 用统计力学中的平均场理论及模拟退火技术,将一般高阶神经网络及玻尔兹曼机的优点结合在一起,以不同于文献[2]的方法,推导出高阶玻尔兹曼机的驰豫动力学的确定性方程和平均场理论学习算法.二者皆便于VLSI实现,且学习算法省去很多CPU时间,对二维镜像对称问题及T—C问题的计算机仿真结果表明三阶玻尔兹曼机的平均场理论学习算法是正确的,且性能较二阶玻尔兹曼机好.更多还原

【Abstract】 Using mean field theory (MFT) of statistical mechanics and simulated annealing technique. the determinate equation of relaxation kinetics of higher-order Boltzmann Machine (BM) and its MFT learning algorithm are deduced in a way different from that in reference[2], which is combined with the merits of general higher-order neural network and Boltzmann Machine. Both are easy to be implemented by VLSI. The learning algorithm saves a lot of CPU time. The computer simulation results for two-dimentional mirror symmetries and T-C problem show that the MFT learning algorithm of the third-order BM is correct and better than that of second-order BM.更多还原