首页> 外国专利> LEARNING PROCEDURES FOR NEURAL NETWORKS BASED ON EVOLUTIONARY ALGORITHMS

LEARNING PROCEDURES FOR NEURAL NETWORKS BASED ON EVOLUTIONARY ALGORITHMS

机译:基于进化算法的神经网络学习程序

摘要

The invention relates to a method for determining weights of a function which maps at least one input value to at least one output value, in such a way that an error is minimized which indicates a deviation of the at least one output value from a target state, the totality of the weights, which forms the function, represents an individual. Starting from a first population, which comprises several individuals, a best individual is determined using a genetic / evolutionary algorithm, with all other individuals in the population being discarded. By varying the individual weights of the best individual, additional individuals are generated which, together with the best individual, form a new population. The new population runs through the genetic / evolutionary algorithm again until at least one predetermined termination condition is met. The variation of the individual weights for determining the individual individuals of the new population takes place on the basis of a predetermined percentage of the individual weights of the best individual. A search for a global extreme of a function can be facilitated by the disclosed method.
机译:本发明涉及一种确定函数的权重的方法,其将至少一个输入值映射到至少一个输出值,以使得误差最小化,这指示从目标状态的至少一个输出值的偏差,重量的整体,形成该功能,代表了个体。从第一个人口开始,其中包括几个人,使用遗传/进化算法确定最好的个体,其中所有其他人都被丢弃。通过改变最好的个人的个人重量,产生额外的个人,以及最好的个人,形成新的人口。新的人口再次通过遗传/进化算法,直到满足至少一个预定的终端条件。用于确定新人人口的个体个人的各个重量的变化是基于最佳个人的个体重量的预定百分比。通过所公开的方法可以促进对功能的全局极端的搜索。

著录项

相似文献

  • 专利
  • 外文文献
  • 中文文献
获取专利

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号