Deep Learning for Natural Language Processeing : Convex Optimization

 效率爆表的一个晚上,只是因为没带手机,可怕!

今天开启新的课程,http://cs224d.stanford.edu/syllabus.html 第一章是凸优化,convex Optimazition

凸集 Convex Set

定义:

A set C is convex if, for any x, y ∈ C and θ ∈ R with 0 ≤ θ ≤ 1,
θx + (1 − θ)y ∈ C.

判别方法:如果一个集合C是凸集,则C中任意两个元素连线上的点都属于C

举例:所有的实数空间;实数空间的非负实数域;

凸方程 Convex Function

定义:定义域D(f)为凸集,且对于任意两个属于D(f)的两个数x,y ; θ ∈ R, 0 ≤ θ ≤ 1,满足
f(θx + (1 − θ)y) ≤ θf(x) + (1 − θ)f(y).

first-order approximation:

 

first order condition for convexity

当且仅当 D(f)是凸集且对于所有满足

则f 是凸方程

second order condition for convexity :

当且仅当 D(f)是凸集且f的海瑟Hessian矩阵(二阶导复合)是半正定:

x ∈ D(f),

Jensen’s Inequality

将凸函数的定义扩展到多个点

若扩展为积分

设定为概率密度

f(E[x]) ≤ E[f(x)]

即为Jensen's Inequality

α-Sublevel Sets

定义:对于凸函数f和α ∈ R,{x ∈ D(f) : f(x) ≤ α}

凸集:f(θx + (1 − θ)y) ≤ θf(x) + (1 − θ)f(y) ≤ θα + (1 − θ)α = α

Convex Optimization Problems

where f is a convex function, gi are convex functions, and hi are affine functions, and x is the optimization variable 

affine function 

optimal value

 

locally optimal if there are no “nearby” feasible points that have a lower objective value

globally optimal if there are no feasible points at all that have a lower objective value 

在凸优化问题中,所有的局部最优都是全局最优

 

凸优化中的特例

Linear Programming

Quadratic Programming

Quadraticallly Constrained Quadratic Programming

Semidefinite Programming

Support Vector Machines 是凸优化中一个典型应用

两类样本中离分类面最近的点且平行于最优分类面的超平面上H1,H2的训练样本就叫做支持向量

问题描述:

假定训练数据 :clip_image008

可以被分为一个超平面:clip_image010

进行归一化:clip_image012

此时分类间隔等于:clip_image014

即使得:最大间隔最大等价于使clip_image016最小

Constrained least squares

Maximum Likelihood for Logistic Regression

minimize ℓ(θ)

应用:

Linear SVM using CVX

posted on 2016-10-16 20:57  Beginnerpatienceless  阅读(293)  评论(0编辑  收藏  举报