fun with hyperplanes 2007

Information about fun with hyperplanes 2007

Published on December 28, 2007

Author: Barbara

Source: authorstream.com

Content

Fun with Hyperplanes: Perceptrons, SVMs, and Friends:  Fun with Hyperplanes: Perceptrons, SVMs, and Friends Ryan Gabbard CIS 391 – Introduction to Artificial Intelligence Universal Machine Learning Diagram:  Universal Machine Learning Diagram Generative v. Discriminative Models:  Generative v. Discriminative Models Generative question: “How can we model the joint distribution of the classes and the features?” Why waste energy on stuff we don’t care about? Let’s optimize the job we’re trying to do directly! Discriminative question: “What features distinguish the classes from one another?” Example:  Example Modeling what sort of bizarre distribution produced these training points is hard, but distinguishing the classes is a piece of cake! chart from MIT tech report #507, Tony Jebara Linear Classification:  Linear Classification Representing Lines:  Representing Lines How do we represent a line? In general a hyperplane is defined by Why bother with this weird representation? Projections:  Projections alternate intuition: recall the dot product of two vectors is simply the product of their lengths and the cosine of the angle between them Now classification is easy!:  Now classification is easy! But... how do we learn this mysterious model vector? Perceptron Learning Algorithm :  Perceptron Learning Algorithm Perceptron Update Example I:  Perceptron Update Example I Perceptron Update Example II:  Perceptron Update Example II Properties of the Simple Perceptron:  Properties of the Simple Perceptron You can prove that if it’s possible to separate the data with a hyperplane (i.e. if it’s linearly separable), the algorithm will converge to that hyperplane. But what if it isn’t? Then perceptron is very unstable and bounces all over the place. Voted Perceptron:  Voted Perceptron Works just like a regular perceptron, except you keep track of all the intermediate models you created. When you want to classify something, you let each of the (many, many) models vote on the answer and take the majority. Properties of Voted Perceptron:  Properties of Voted Perceptron Simple! Much better generalization performance than regular perceptron (almost as good as SVMs) Can use the ‘kernel trick’ Training as fast as regular perceptron But run-time is slower. Averaged Perceptron:  Averaged Perceptron Extremely simple! Return as your final model the average of all your intermediate models. Approximation to voted perceptron. Nearly as fast to train and exactly as fast to run as regular perceptron. What’s wrong with these hyperplanes?:  What’s wrong with these hyperplanes? They’re unjustifiably biased!:  They’re unjustifiably biased! A less biased choice:  A less biased choice Margin:  Margin The margin is the distance to closest point in the training data. We tend to get better generalization to unseen data if we choose the separating hyperplane which maximizes the margin. Support Vector Machines:  Support Vector Machines Another learning method which explicitly calculates the maximum margin hyperplane by solving a gigantic quadratic programming problem. Generally considered the highest-performing current machine learning technique. But it’s relatively slow and very complicated. Margin-Infused Relaxed Algorithm (MIRA):  Margin-Infused Relaxed Algorithm (MIRA) Multiclass; each class has a prototype vector Classify an instance by choosing the class whose prototype vector has the greatest dot product with the instance During training, when updating make the ‘smallest’ (in a sense) change to the prototype vectors which guarantees correct classification by a minimum margin Pays attention to the margin directly What if it isn’t separable?:  What if it isn’t separable? Project it to someplace where it is!:  Project it to someplace where it is! Kernel Trick:  If our data isn’t linearly separable, we can define a projection to map it into a much higher dimensional feature space where it is. For some algorithms where everything can be expressed as the dot products of instances (SVM, voted perceptron, MIRA) this can be done efficiently using something called the `kernel trick’ Kernel Trick

Related presentations


Other presentations created by Barbara

Solar System
17. 06. 2007
0 views

Solar System

Advanced SQL Injection
30. 08. 2007
0 views

Advanced SQL Injection

PrivateExchange
22. 04. 2008
0 views

PrivateExchange

07 fordjob1
17. 04. 2008
0 views

07 fordjob1

20061011114434853
13. 04. 2008
0 views

20061011114434853

Bruce Lambert Army Corps
10. 04. 2008
0 views

Bruce Lambert Army Corps

SPAC2007 Juan Rodriguez
09. 04. 2008
0 views

SPAC2007 Juan Rodriguez

Chapter7
07. 04. 2008
0 views

Chapter7

tourism chapter 04
30. 03. 2008
0 views

tourism chapter 04

LAC International Trade
28. 03. 2008
0 views

LAC International Trade

feb2006final
27. 03. 2008
0 views

feb2006final

virtualcommunities
26. 03. 2008
0 views

virtualcommunities

Mickey Mouse
26. 06. 2007
0 views

Mickey Mouse

1who gets tb in nyc
27. 09. 2007
0 views

1who gets tb in nyc

lijian
12. 10. 2007
0 views

lijian

O2 Diesel
08. 11. 2007
0 views

O2 Diesel

American Romanticism
30. 08. 2007
0 views

American Romanticism

233nm60
30. 08. 2007
0 views

233nm60

MBA Lecture Series v2
30. 08. 2007
0 views

MBA Lecture Series v2

hep2005 talk MarkVagins
09. 10. 2007
0 views

hep2005 talk MarkVagins

Control Tech
05. 12. 2007
0 views

Control Tech

DasuCMSTriggerUCSD
07. 10. 2007
0 views

DasuCMSTriggerUCSD

ams ppt
30. 08. 2007
0 views

ams ppt

Question Answering
16. 11. 2007
0 views

Question Answering

Facts x about Finland
22. 11. 2007
0 views

Facts x about Finland

OWAS PAppSecEU2006 CLASP Project
30. 08. 2007
0 views

OWAS PAppSecEU2006 CLASP Project

OWASP Flyer Sep06
30. 08. 2007
0 views

OWASP Flyer Sep06

american history
28. 12. 2007
0 views

american history

Frank Garber Presentation
02. 01. 2008
0 views

Frank Garber Presentation

DPS07 65 01 Fritzius
03. 01. 2008
0 views

DPS07 65 01 Fritzius

Teaching Political Sociology
04. 01. 2008
0 views

Teaching Political Sociology

Gaming in Education
07. 01. 2008
0 views

Gaming in Education

Plume tracking hardware
07. 01. 2008
0 views

Plume tracking hardware

Altera
28. 11. 2007
0 views

Altera

dead reckon cdr
07. 01. 2008
0 views

dead reckon cdr

Infections 3
04. 12. 2007
0 views

Infections 3

CMC IR1001
27. 09. 2007
0 views

CMC IR1001

class2 3
16. 11. 2007
0 views

class2 3

mixload
06. 11. 2007
0 views

mixload

web query 0609
07. 11. 2007
0 views

web query 0609

FSA
27. 12. 2007
0 views

FSA

CompanyDossier
29. 09. 2007
0 views

CompanyDossier

Hunting For Black Holes
28. 11. 2007
0 views

Hunting For Black Holes

DAR
20. 02. 2008
0 views

DAR

8 Soci 1015 Chapter7 Family
24. 02. 2008
0 views

8 Soci 1015 Chapter7 Family

ABSSEI Oswald
29. 02. 2008
0 views

ABSSEI Oswald

NeMO Curr Part3 v2
26. 06. 2007
0 views

NeMO Curr Part3 v2

nelson sheinberg Presentation
26. 06. 2007
0 views

nelson sheinberg Presentation

n0002 SPIE1
26. 06. 2007
0 views

n0002 SPIE1

Metric System 1
26. 06. 2007
0 views

Metric System 1

media kit
26. 06. 2007
0 views

media kit

March 14 PMI Presentation
26. 06. 2007
0 views

March 14 PMI Presentation

fountain of age
26. 06. 2007
0 views

fountain of age

Lifting Equation
13. 12. 2007
0 views

Lifting Equation

Dietary Guidelines
04. 03. 2008
0 views

Dietary Guidelines

upshur pc1
10. 03. 2008
0 views

upshur pc1

crossref
30. 08. 2007
0 views

crossref

ddbppt
20. 11. 2007
0 views

ddbppt

DEPBasicsCourse
30. 12. 2007
0 views

DEPBasicsCourse

guerra
12. 11. 2007
0 views

guerra

James F Cooper
30. 08. 2007
0 views

James F Cooper

lubin talk
03. 01. 2008
0 views

lubin talk

NDD presentation compressed
30. 08. 2007
0 views

NDD presentation compressed

madcooper
07. 12. 2007
0 views

madcooper

graduacion1
01. 01. 2008
0 views

graduacion1

GBIF demo Japan081003
27. 11. 2007
0 views

GBIF demo Japan081003

20061019 1732 oberauer hql06
15. 11. 2007
0 views

20061019 1732 oberauer hql06

phpulse oct
05. 01. 2008
0 views

phpulse oct

media searching
26. 06. 2007
0 views

media searching

Smith Core values
17. 06. 2007
0 views

Smith Core values

Smith1
17. 06. 2007
0 views

Smith1

Significance of the Cross
17. 06. 2007
0 views

Significance of the Cross

Sharp
17. 06. 2007
0 views

Sharp

section 2 attitude to food
17. 06. 2007
0 views

section 2 attitude to food

Section4 5
17. 06. 2007
0 views

Section4 5

Spirituality
17. 06. 2007
0 views

Spirituality

sonnet presentation
17. 06. 2007
0 views

sonnet presentation

Star addition tutorial
17. 06. 2007
0 views

Star addition tutorial

stand up comedy
17. 06. 2007
0 views

stand up comedy

SS 1SBrown
17. 06. 2007
0 views

SS 1SBrown

Emerson Transcendentalism
30. 08. 2007
0 views

Emerson Transcendentalism

ABinEurope
23. 11. 2007
0 views

ABinEurope

TextMining 06
03. 10. 2007
0 views

TextMining 06

oct04ach
05. 11. 2007
0 views

oct04ach

SCP2
17. 06. 2007
0 views

SCP2

transcendentalism
30. 08. 2007
0 views

transcendentalism

micro ch03 presentation
04. 10. 2007
0 views

micro ch03 presentation

SC morning
17. 06. 2007
0 views

SC morning

ISIC cobrandNEUenglish
18. 03. 2008
0 views

ISIC cobrandNEUenglish

02b LisbonWeb
30. 12. 2007
0 views

02b LisbonWeb

ProvenceArchitecture
05. 11. 2007
0 views

ProvenceArchitecture

san diego 04
01. 11. 2007
0 views

san diego 04

noemie 2
26. 06. 2007
0 views

noemie 2

Community Service PP 06 FOR WEB
05. 11. 2007
0 views

Community Service PP 06 FOR WEB

Sections3 7
17. 06. 2007
0 views

Sections3 7

ECE TRANS WP29 GRSP 41 inf09e
26. 11. 2007
0 views

ECE TRANS WP29 GRSP 41 inf09e

srwg graz
26. 11. 2007
0 views

srwg graz

Meydan
23. 11. 2007
0 views

Meydan

LWS05
02. 11. 2007
0 views

LWS05

mal 2005 bra
30. 08. 2007
0 views

mal 2005 bra

Standards Aligned Classroom
17. 06. 2007
0 views

Standards Aligned Classroom

steenkampNVDRS
06. 03. 2008
0 views

steenkampNVDRS