Category Archives: 机器学习

【MLT】12-Neural Network

本章是神经网络的入门,包含如下四个方面:

  1. Motivation:基本单位-神经元,与生物神经元的联系
  2. Neural Network Hypothesis:神经元的转换函数(激活函数),物理解释
  3. Neural Network Learning:网络权重的学习,网络权重的偏微分推导,反向传播(BP)
  4. Optimization and Regularization:网络优化,模型VC维,两种regularization方法(正则化项,early stopping)
 

Continue reading

Support Vector Machines vs Artificial Neural Networks

svms.org网站上一篇神经网络和支持向量机的对比文章,这里做简单翻译,如有错误请在下方评论处留言,感谢作者!感谢爱可可-爱生活分享!

 

Continue reading

【MLF】2-4-Non-Separable Data

Lecture 2: Learning to Answer Yes/No——Non-Separable Data

本节重点介绍了线性不可分数据的处理办法——PLA算法的改进算法——口袋算法(Pocket Algorithm)。

  • Content
  • Slide #1. More about PLA
  • Slide #2. Learning with Noisy Data
  • Slide #3. Line with Noise Tolerance
  • Slide #4. Pocket Algorithm
  • Slide #5. Fun Time
  • Slide #6. Summary

Continue reading

【MLF】2-3-Guarantee of PLA

Lecture 2: Learning to Answer Yes/No——Guarantee of PLA

Slide #1. Linear Separability

我们来看看PLA(Perceptron Learning Algorithm)什么时候会停下来,在想这个问题之前不妨想想PLA的终止条件是什么,是PLA找到一条线能毫无错误地将数据样本分开,但是这有一个前提条件,就是数据样本可以用一条线分开,否则PLA永远无法将数据样本分开。

QQ截图20150606183755

 

 

 

 

 

 

Continue reading

【MLF】2-2-Perceptron Learning Algorithm (PLA)

 Lecture 2: Learning to Answer Yes/No——Perceptron Learning Algorithm (PLA)

通过上节课我们知道了一个可能的Hypothesis长相(可能是一条线,或者是在空间里所有可能的高维平面),那么我们现在的问题是——我们要怎样设计一个算法从众多的线或者面里选择一条最好的出来。

不妨我们想想,我们认为最好的线是什么,理想上的f对吧?如果能算出理想的f,那么我们就没什么好做的了,因为我们不知道f,所以我们希望我们的g和f越接近越好。但我们知道我们的f是由我们的数据产生的。  Continue reading

【MLF】2-1-Perception Hypothesis Set

 Lecture 2: Learning to Answer Yes/No——Perception Hypothesis Set

本节是第二讲(Lecture 2: Learning to Answer Yes/No)的第一节课,主要是对感知器(本课第一个演算法)的入门。首先我们回顾了一下上一讲(Lecture 1: The Learning Problem),同样是通过信用卡检测的问题引入本节,机器学习演算法如何做判断是非(Learning to Answer Yes/No)的问题。

Slide #1. Roadmap

  • When Can Machines Learning?
  • ——Lecture 1: The Learning Problem
  • ——A takes D and H to get g
  • ————Course Introductionfoundation oriented and story-like
  • ————What is Machine Learninguse data to approximate target
  • ————Application of Machine Learningalmost everywhere
  • ————Components of Machine Learning: A takes D and H to get g
  • ————Machine Learning and Other Fieldrelated to DM, AI and Stats
  • ——Lecture 2: Learning to Answer Yes/No
  • ————Perception Hypothesis Set
  • ————Perception Learning Algorithm (PLA)
  • ————Guarantee of PLA
  • ————Non-Separable Data
  • ②Why Can Machines Learn?
  • ③How Can Machines Learn?
  • ④How Can Machines Learn Better?

Continue reading

【MLF】1-5-Machine Learning and Other Fields

Lecture 1: The Learning Problem——Machine Learning and Other Fields

本节主要讲了ML和DM(二者难以区分)ML和AI(ML是实现AI的方法之一),以及ML和Stats(Stats是实现ML的方法之一)的关系。最后对本章(第一章)进行了总结。

  • difficult to distinguish ML and DM in reality
  • ML is one possible route to realize AI
  • statistics: many useful tools for ML

Continue reading

【MLF】1-4-Components of Machine Learning

Lecture 1: The Learning Problem——Components of Machine Learning

本节课从一个例子(信用卡检测)说起,分别介绍了机器学习中的五个重要(unknown target function, training examples, learning algorithm, hypothesis set, final hypothesis)的部分(如下图)。machine learning: use data to compute hypothesis g that approximates target f。

Slide #1. Components of Learning: Metaphr Using Credit Approval

Application Information

  • age: 23 years
  • gender: female
  • annual salary: NTD 1,000,000
  • year in residence: 1 year
  • year in job: 0.5 year
  • current debt: 200,000

unknown pattern to be learned: ‘approve credit card good for bank’ Continue reading

【MLF】1-3-Applications of Machine Learning

Lecture 1: The Learning Problem——Applications of Machine Learning

本节课主要谈了下ML可以应用的领域(除衣食住行外,还有教育,娱乐等等)。

Slide #1. Daily Needs: Food, Clothing, Housing, Transportation

data ——》ML——》skill

  1. Food (Sadilek et al., 2013):①data: Twitter data (words + location)②skill: tell food poisoning likeliness of restaurant properly
  2. Clothing (Abu-Mostafa, 2012):①data: sales figures + client surveys②skill: give good fashion recommodations to clients
  3. Housing (Tsanas and Xifara, 2012):①data:characteristics of buildings and their energy load②skill: predict energy load of other buildings closely
  4. Transportation (Stallkamp et al,. 2012):①data: some traffic sign images and meanings②skill: recognize traffic signs accurately

ML is everywhere! Continue reading

【MLF】1-2-What is Machine Learning?

Lecture 1: The Learning Problem——What is Machine Learning?

本节主要讲了什么是ML,从一个识别树的例子入手。比较重要的是决定什么时候使用ML的三个关键(key essence):

  1. exists some ‘underlying pattern’ to be learned——so ‘performance mearsure’ can be improved
  2. but no programmable(easy) definition——so ‘ML’ is needed
  3. somehow there is data about the pattern——so ML has some ‘inputs’ to learn from

Slide #1. From Learning to Machine Learning

  • learning: acquiring skill with experience accumulated from observations.
  • obersvations =》 Learning =》 skill
  • machine learning: acquiring skill with experience accumulated/computed from data.
  • data =》ML=》skill

Continue reading