基于同伦方程的交叉熵BP算法

摘要

提出一种建构在同伦方程基础上的交叉熵BP算法,在原有的交叉熵函数基础上,通过同伦方程的参数逐步调节权值,使收敛达到最佳效果.同时其激活函数采用广泛意义的函数,有利于拓展函数使用空间,并结合动量项使收敛效果更佳.实验结果表明,改进后的算法与原始的加动量项算法相比有较好的收敛速度,陷入极小点的几率也大为降低.%A cross entropy BP algorithm built on homotopy equation is put forward. On the basis of the original cross entropy function, its weight can be gradually adjusted by homotopy equation parameters so that the convergence can achieve the foremost effect. Meanwhile its activation function is chosen for broad sense to help with expanding the space used by the function, additionally by integrating add-momentum to achieve even better convergence effect. Experiment results show that, compared to the original add-momentum algorithm, the improved algorithm is faster at convergence and more resistant against minimum point traps.