下降方向

在CDMA系统中实现多用户检测的障碍函数下降方向法

Barrier function descent direction method for CDMA multi-user detection

这种方法每步迭代不用求解QP子问题，而是求解线性方程组来得到可行下降方向。

At each iteration , the methods only need to solve linear equation systems in stead of QP subproblem to obtain feasible and descent direction .

关于线性等式约束极小化中的ABS下降方向

On ABS descent directions for constrained minimization with linear equality constraints

然后给出以δX为下降方向的一族含双参数的下降函数，并且在一定条件下证明了该函数族关于结构拟牛顿法产生的点列是严格下降的。

Specifically , the paper presents a family of descent functions containing two parameters and proves that the functions are strictly descent about sequence of points produced by the structured quasi Newton method under some conditions .

对于多层前馈网络的权值的计算，人们多采用基于梯度下降方向搜索的误差反向传播（EBP）算法。

People mainly apply the back propagation algorithm ( BP algorithm ) based on gradient descent direction of search to train the weights of the multi-layered feedforward neural network .

在本文中，我们首先介绍一种动量项系数根据目前的梯度最速下降方向（负梯度方向）和权值的上一步改变的方向的信息而自适应变化的BPM算法。

In this thesis , our first work is to introduce an efficient BP-Momentum algorithm , where the momentum coefficient is adjusted dynamically based on the information about the current gradient and the weight change of previous step at each cycle of training procedure .

尽管如此，只有当目前的梯度下降方向与上一步的权值改变的方向相近时，这种动量项系数为正常数的BPM算法才具备加速的功能。

However , with a fixed momentum parameter , the momentum term will have an accelerating effect only if the current downhill gradient of the error function & E_ ω( k ) and the last change in weight △ω( k-1 ) has a similar direction .

只需要作矩阵的乘法运算或求解方程组就可以得到minimax的下降方向.该算法具有全局收敛性，数值例子表明，该方法具有良好的数值计算结果。

The advantage of this algorithm is that only making the matrix multiplication operation or solving equations set , can the decending direction of minimax be obtained without having need to consider how many effective functions there are and how to calculate the inverse matrix .

算法引入了一个新的Armijo型步长搜索，它可从任意初始点开始，而且在有限步之内可产生一个可行点，之后自动变为一个可行下降方向算法。

The algorithm , in which a new Armijo type step-length search is introduced , starts with an arbitrary initial point , and it must generate a feasible point in a finite number of iterations , then automatically becomes a feasible descent SQP algorithm .

集值函数的下降方向锥的一个性质

A Characterization of Decrease Direction Cone with Respect to Vector-valued Function

提出了使用变分导数求解泛函的下降方向。

Make use of functional derivative for solving functional decline direction .

可以证明这样的投影方向是可行下降方向。在一定条件下，我们证明了所给算法的全局收敛性及超线性收敛性。

It 's global convergence and the superlinear convergence are proved under suitable conditions .

人工势场在局部极大值和最速下降方向方面的特殊性质进一步优化了算法。

The algorithm is optimized using the local maximum value of the potential and the steepest descent direction .

本文算法同时考虑了步长与下降方向。

In this paper , our algorithm is set up by considering both search direction and step size .

在每次迭代中，通过求解一个二次子规划得到一个可行下降方向。

At each iteration , a feasible direction of descent is obtained by solving a quadratic programming ( QP ) .

修改后的方法在算法进行过程中能够产生下降方向，算法无需重开始。

The modified schemes generate descent directions at each iteration and thus there is no need to restart the algorithm .

其次，我们研究原张量模型在其非稳定点下降方向的选取。

Third , we consider the methods to select a descent direction of the original tensor model at its non-critical point .

带下降方向搜索的总极值方法积分总极值方法在有界可测函数上的推广

An Algorithm for Finding Global Minimun with Descent Direction Extension of Integral-Level Set Method for Global Optimization to Bounded Measurable Functions

在每次迭代时，这种方法都利用一种线性约束半正定二次问题来产生一个合理的下降方向；

At each iteration , the method uses a linearly constrained positive semidefinite quadratic problem to generate a feasible descent direction .

算法通过解系列二次规划寻找可行下降方向。新的迭代点由不精确线搜索产生。

The method finds the feasible descent directions by solving successive quadratic programming , and obtains new iteration points by inexact line search .

有限次迭代后，迭代点进入可行集且主方向是一可行下降方向。

After finite iterations , the iteration point gets into the feasible set and the master direction is a feasible direction of descent .

该方法对给定的参数在下降方向寻找逼近函数的最优点。

For a given value of the parameter , the method searches for an optimum point of the approximate function in a descent direction .

在盲解卷模型的参数空间里，代表代价函数最陡下降方向的不是传统的梯度而是自然梯度。

The natural gradient rather than conventional one gives the steepest descent direction of loss function in the parameter space of blind source separation .

该算法的优点是：（1）在无需线性搜索的条件下，迭代方向就是充分下降方向；

The method has the following attractive properties : ( 1 ) the iterative direction is always a sufficiently descent direction without utilizing the line search ;

本文给出一种计算步长的方法，此方法的优点为：若在此下降方向上解存在，那么新方法以较少的计算量确定解的存在区间（基于0－618法）；

( I ) if there exist solutions in this decrease direction , then we can find a solution interval ( based on 0 618 method );

与传统方法以及神经网络优化算法相比较，作者提出的方法是利用障碍函数而得到的一个新的下降方向而不是梯度下降方向。

As compared to conventional detectors and neural network optimal algorithms , the method derives a new descent direction from the barrier function other than the gradient direction .

文章给出了一种新的非精确线性搜索下的共轭梯度法，说明了在新线性搜索下每次迭代能够产生下降方向。

A new conjugate gradient algorithm with a new inexact line search is gived , the descent directions has produced in each iterative with the inexact line search .

首先，从二次函数在一点的最速下降方向出发定义二次函数过一点的最速下降曲线；

In this paper , the steepest descent curve of quadratic function through one point was first defined from the steepest descent direction of quadratic function at one point .

这样在每一个非稳定点处至少存在一个可行的下降方向，在初始点是可行点的条件下，此算法产生的每一个迭代点都是可行点。

So at every nonstationary point there exists at least one feasible descent direction , every iterate generated by the methods is feasible so long as the initial iterate is feasible .

根据原优化问题的二次近似模型，运用拟牛顿方向与最速下降方向之凸组合作为搜索方向，采用了新的策略。

According to the quadratic approximate model of the original optimization problem , the trust region algorithm uses directions , a convex combination of the quasi-newton direction and the steepest descent direction .