然而,变步长自然梯度算法性能仍然不够理想。
However, the performance of variable step-size natural gradient algorithm is still less than ideal.
所以我们又给出一种自适应修正自然梯度算法,用一个合适的距离测度函数来控制步长和动力因子。
So we present an adaptive improved natural gradient algorithm, which use an appropriate estimation function to control the step-size and the momentum factor.
为了解决自然梯度算法的这个缺点,人们提出了变步长自然梯度算法,既可以获得较快的收敛速度,又可以减小稳定误差。
Variable step-size natural gradient algorithm which enjoys faster convergence speed and smaller stability error is proposed to solve the shortcoming of natural gradient algorithm.
应用推荐