self-cleaning gradient

Aiming at the defect of gradient-descending arithmetic, such as low-speed of constringency, easily falling into local minimum, an improved approach based on affixation momentum and self-adaptive learning rate is proposed.

针对传统BP算法即梯度下降法，收敛速度慢、容易陷入局部极小值等缺点，提出了基于附加动量法和自适应学习速率的改进方法。

Training networks combining genetic algorithm ( GA) based on global optimization with backpropagation ( BP) based on gradient descent in the paper make the linking weights of networks self-adaptive evolution in constantly iterative process.

本文采用全局寻优的遗传算法（GA）和基于梯度下降的局部寻优反传算法（BP）相结合来训练网络，使网络的连接权在不断迭代过程中自适应演化。

Taking into consideration of the effect of self-shading, it can effectively estimate the control parameters used in the SFS algorithms. It implements the brightness constraint, the intensity gradient constraint and the integrability constraint to compute the surface height and orientation and to reconstruct the surface shape.

它在考虑自遮掩影响的情况下，有效地估计了SFS算法中涉及的各种控制参数，并引入亮度约束、灰度梯度约束和可积性约束，计算出表面高度和表面向量，实现三维重构。

A reverse transmission calculation of self adapting learning rate with momentum gradient reduction is described in the paper against the problems of slow convergence and easy trapping into smallest spot.

针对BP网络学习收敛速度慢和易陷入局部最小点的不足，提出了一种自适应学习速率动量梯度下降反向传播算法对BP神经网络进行训练。

Especially back propagation algorithm based on self adapting learning rate with momentum gradient reduction can avoid efficiently the slow convergence and the problem of 'local infinitesimal value' of BP network.

其中自适应学习速率动量梯度下降反向传播算法，可以有效避免BP网络收敛速度慢和存在所谓局部最小值问题。