Deep Learning for Natural Language Processeing : Convex Optimization

Posted

tags:

篇首语:本文由小常识网(cha138.com)小编为大家整理,主要介绍了Deep Learning for Natural Language Processeing : Convex Optimization相关的知识,希望对你有一定的参考价值。

 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:

 技术分享

first order condition for convexity

当且仅当 D(f)是凸集且对于所有技术分享满足

技术分享

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),

技术分享

Jensen’s Inequality

将凸函数的定义扩展到多个点

技术分享

若扩展为积分

技术分享

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

技术分享

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

affine function 技术分享

optimal value

技术分享

 

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

技术分享

Quadratic Programming

技术分享

Quadraticallly Constrained Quadratic Programming

技术分享

Semidefinite Programming

技术分享

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 :技术分享

可以被分为一个超平面:技术分享

进行归一化:技术分享

此时分类间隔等于:技术分享

即使得:最大间隔最大等价于使技术分享最小

技术分享

Constrained least squares

技术分享

Maximum Likelihood for Logistic Regression

技术分享

minimize ?(θ)

应用:

Linear SVM using CVX



以上是关于Deep Learning for Natural Language Processeing : Convex Optimization的主要内容,如果未能解决你的问题,请参考以下文章

NOTE:Deep Reinforcement Learning with a Natural Language Action Space

NOTE:Deep Reinforcement Learning with a Natural Language Action Space

Coursera, Deep Learning 5, Sequence Models, week2, Natural Language Processing & Word Embeddings

TensorFlow by Google Machine Learning Foundations: Ep #8 - Tokenization for Natural Language Process

Deep_Learning_for_Computer_Vision_with_Python.pdf

Joint Deep Learning for Pedestrian Detection笔记