2011 Matrix optimization - searching between the first and second order methods (slide).pdf
- 1、本文档共37页,可阅读全部内容。
- 2、有哪些信誉好的足球投注网站(book118)网站文档一经付费(服务费),不意味着购买了该文档的版权,仅供个人/单位学习、研究之用,不得用于商业用途,未经授权,严禁复制、发行、汇编、翻译或者网络传播等,侵权必究。
- 3、本站所有内容均由合作方或网友上传,本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺!文档内容仅供研究参考,付费前请自行鉴别。如您付费,意味着您自己接受本站规则且自行承担风险,本站不退款、不进行额外附加服务;查看《如何避免下载的几个坑》。如果您已付费下载过本站文档,您可以点击 这里二次下载。
- 4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等,请点击“版权申诉”(推荐),也可以打举报电话:400-050-0827(电话支持时间:9:00-18:30)。
查看更多
2011 Matrix optimization - searching between the first and second order methods (slide)
bb
b
b
b
b
b
b
b
b
b
b
b
bb b
b b
b
b
b
b
b
b
b
b
b
b
b b
b
b
b
b
b
Matrix Optimization: Searching between
the First and Second Order Methods
Defeng Sun
Department of Mathematics and Risk Management Institute
National University of Singapore
May 17, 2011
The Matrix Optimization Problem
The SIAM Conference on Optimization 2011, Darmstadt NUS/SUN – 2 / 37
We consider the “standard” matrix optimization problem (MOP) and its
dual:
(P)
min 〈c, x〉+ f(x)
s.t. Ax = b
and
(D)
max 〈b, y〉 ? f?(z)
s.t. A?y ? c = z ,
where X is the Cartesian product of several finite dimensional real matrix
spaces, symmetric or non-symmetric,
A? is the adjoint of the linear operator A : X → ?m, c ∈ X , b ∈ ?m,
f : X → (?∞,∞] is a closed proper convex function with its Fenchel
conjugate f?.
The SIAM Conference on Optimization 2011, Darmstadt NUS/SUN – 3 / 37
The Fenchel conjugate of f is defined by
f?(z) := sup
x∈X
{〈z, x〉 ? f(x)} .
In standard linear programming, f(x) = δ?n
+
(x), the indicator function
over ?n+ and f
?(x) = δ(??n
+
)(x).
In semidefinite programming (SDP), f(x) = δSn
+
, the indicator function
over Sn+ and f
?(x) = δ(?Sn
+
)(x).
Desirable Properties of f
The SIAM Conference on Optimization 2011, Darmstadt NUS/SUN – 4 / 37
We need conditions on f . Specifically, we require
? The Moreau-Yosida regularization of f
ψf (x) := min
z∈X
{
f(z) +
1
2
‖z ? x‖2
}
has a closed form solution, denoted by Pf (x).
? We can easily compute the directional derivative of
?ψf(x) = x? Pf(x).
? The function ?ψf is (strongly) semismooth.
The SIAM Conference on Optimization 2011, Darmstadt NUS/SUN – 5 / 37
Let us first look at one simple example with nonsymmetric matrices:
min
y∈?k
‖A0 ?
∑k
i=1 yiAi‖2 , (1)
where Ai are m by n matrices, ‖ · ‖2 is the spectral (operator) norm of
matrices (the largest singular value).
Use ‖ · ‖? to denote the nuclear norm (the sum of all singular values) and
B1? to denote the unit nuclear norm ball.
The SIAM Conference on Optimization 2011, Darmstadt NUS/SUN – 6 / 37
We c
您可能关注的文档
- 11个雅思听力评分标准整理.pdf
- 11期外刊班(第一、二天)讲义.pdf
- 11年专升本计科类英语.pdf
- 11款不泛油光遮盖力强粉底推荐.pdf
- 12 I can swim fast.1.ppt
- 12 不确定性 uncertainty 2009-11-02.ppt
- 12 hcm2000 HIGHWAY CONCEPTS.pdf
- 120220-0210;120220-0161;120220-0202;120220-0204;120220-0206;中文规格书,Datasheet资料.pdf
- 12款控油乳液 夏日肌肤无油.pdf
- 137Cs and 210Pb 在土壤侵蚀中的应用.pdf
- 七章货物的保险.pptx
- 三章国际间接投资.pptx
- 人性假设理论.pptx
- 外研高一英语必修三ModuleIntroduction汇总市公开课获奖课件省名师示范课获奖课件.pptx
- 月相成因优质获奖课件.pptx
- 小学二年级语文课件《狐假虎威》省名师优质课赛课获奖课件市赛课一等奖课件.pptx
- 养羊业概况专题知识讲座.pptx
- 微生物的实验室培养市公开课获奖课件省名师示范课获奖课件.pptx
- 人教版六年级下册式与方程整理与复习市公开课获奖课件省名师示范课获奖课件.pptx
- 必威体育精装版高中精品语文教学:第二单元-第7课-诗三首:涉江采芙蓉、-短歌行、归园田居市公开课获奖课件省名师.pptx
文档评论(0)