图 书 馆
第四讲:对偶问题的投影次梯度Lecture 4: Project Subgradient For Dual Problem |
|
| 课程网址: | http://videolectures.net/stanfordee364bs08_boyd_lec04/ |
| 主讲教师: | Stephen P. Boyd |
| 开课单位: | 斯坦福大学 |
| 开课时间: | 2010-06-21 |
| 课程语种: | 英语 |
| 中文简介: |  当然，如果严格可行，我们将需要强对偶性。我们将拥有 Slater 的条件，并且强二元性将成立。这给了你零对偶差距，我想如果你没有那个，那么你根本无法解决这个问题，因为最优值甚至不一样。所以让我们假设。实际上，还有更多。大锤条件是这样的。你需要的是，当你找到 lambda* 时，你想要的是 lambda* 处的拉格朗日函数在 x 中应该有一个唯一的极小值。如果是这样，那么 x 实际上就是这里的 x*。好的？所以这就是条件。 ... |
| 课程简介: | Sure, we’re going to need strong duality holds if it were strictly feasible. We’d have Slater’s condition and strong duality would hold. That gives you zero duality gap and I guess if you don’t have that, then you can’t solve this at all, because the optimal values aren’t even the same. So let’s assume that. There’s more, actually, to it than just that. What the sledgehammer condition is is this. What you’ll need is that when you find lambda*, what you want is that the Lagrangian at lambda* should have a unique minimizer in x. If it does, then that x is actually x* up here. Okay? So that’s the condition. ... |
| 关 键 词: | Slater; 零对偶差距; 拉格朗日函数 |
| 课程来源: | 视频讲座网 |
| 数据采集: | 2021-07-17：nkq |
| 最后编审: | 2021-08-28：nkq |
| 阅读次数: | 84 |