j*****n
发帖数: 1545 | 1 【 以下文字转载自 CS 讨论区 】
发信人: jetchen (飞机), 信区: CS
标 题: 急问个优化的问题
发信站: BBS 未名空间站 (Fri Jan 14 16:43:03 2011, 美东)
有一个functon f(x,y), 要找optimum. 直接优化x和y难度很大, 但是可以分别优化:
固定y, 优化x; 固定x, 优化y. 而且这个分别优化能够保证是 global optimum.
我的问题是我的这个迭代的算法能保证最后的结果是最优吗?
step 1: fix y_0, update x_0 to the optimal solution x_1
step 2: fix x_1, update y_1 to the optimal solution y_2
step 3: go to step 1 and iterate until convergence.
我觉得好像和conjugate gradient系列的算法很像, 但我一下想不出来我这个算法是对
还是错. 求指点. 如果有定理之类的可以参考最好！！ | s*******g
发帖数: 483 | 2 Only when the function is jointly convex with respect to both x and y could
one expect global optimum via alternating optimization | j*****n
发帖数: 1545 | 3 Thanks, someone point me to the field of bi-convex programming.
could
【在 s*******g 的大作中提到】
: Only when the function is jointly convex with respect to both x and y could
: one expect global optimum via alternating optimization
| s*******g
发帖数: 483 | 4 Only when the function is jointly convex with respect to both x and y could
one expect global optimum via alternating optimization |
|
