INTRODUCTION TO MACHINE LEARNING 3RD EDITION

Similar documents
Lecture Slides for INTRODUCTION TO. Machine Learning. ETHEM ALPAYDIN The MIT Press,

CHAPTER 7: CLUSTERING

Clustering (Bishop ch 9)

Normal Random Variable and its discriminant functions

Lecture Slides for INTRODUCTION TO. Machine Learning. ETHEM ALPAYDIN The MIT Press,

GMM parameter estimation. Xiaoye Lu CMPS290c Final Project

CS 536: Machine Learning. Nonparametric Density Estimation Unsupervised Learning - Clustering

THEORETICAL AUTOCORRELATIONS. ) if often denoted by γ. Note that

Learning Objectives. Self Organization Map. Hamming Distance(1/5) Introduction. Hamming Distance(3/5) Hamming Distance(2/5) 15/04/2015

Lecture Slides for INTRODUCTION TO. Machine Learning. ETHEM ALPAYDIN The MIT Press,

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 4

Machine Learning 2nd Edition

Conservation of Momentum. The purpose of this experiment is to verify the conservation of momentum in two dimensions.

CHAPTER 10: LINEAR DISCRIMINATION

FI 3103 Quantum Physics

( ) [ ] MAP Decision Rule

Discrete Markov Process. Introduction. Example: Balls and Urns. Stochastic Automaton. INTRODUCTION TO Machine Learning 3rd Edition

LECTURE :FACTOR ANALYSIS

1 Widrow-Hoff Algorithm

EP2200 Queuing theory and teletraffic systems. 3rd lecture Markov chains Birth-death process - Poisson process. Viktoria Fodor KTH EES

Machine Learning. Lecture Slides for. ETHEM ALPAYDIN The MIT Press, h1p://

Sklar: Sections (4.4.2 is not covered).

A New Method for Computing EM Algorithm Parameters in Speaker Identification Using Gaussian Mixture Models

Clustering with Gaussian Mixtures

Bayes rule for a classification problem INF Discriminant functions for the normal density. Euclidean distance. Mahalanobis distance

Solution in semi infinite diffusion couples (error function analysis)

Chapter 6 DETECTION AND ESTIMATION: Model of digital communication system. Fundamental issues in digital communications are

John Geweke a and Gianni Amisano b a Departments of Economics and Statistics, University of Iowa, USA b European Central Bank, Frankfurt, Germany

( ) () we define the interaction representation by the unitary transformation () = ()

Viscous Damping Summary Sheet No Damping Case: Damped behaviour depends on the relative size of ω o and b/2m 3 Cases: 1.

Math 128b Project. Jude Yuen

Response of MDOF systems

Advanced Machine Learning & Perception

OP = OO' + Ut + Vn + Wb. Material We Will Cover Today. Computer Vision Lecture 3. Multi-view Geometry I. Amnon Shashua

Outline. Probabilistic Model Learning. Probabilistic Model Learning. Probabilistic Model for Time-series Data: Hidden Markov Model

V.Abramov - FURTHER ANALYSIS OF CONFIDENCE INTERVALS FOR LARGE CLIENT/SERVER COMPUTER NETWORKS

Chapter Lagrangian Interpolation

CS286.2 Lecture 14: Quantum de Finetti Theorems II

10. A.C CIRCUITS. Theoretically current grows to maximum value after infinite time. But practically it grows to maximum after 5τ. Decay of current :

Department of Economics University of Toronto

In the complete model, these slopes are ANALYSIS OF VARIANCE FOR THE COMPLETE TWO-WAY MODEL. (! i+1 -! i ) + [(!") i+1,q - [(!

Foundations of State Estimation Part II

II The Z Transform. Topics to be covered. 1. Introduction. 2. The Z transform. 3. Z transforms of elementary functions


Graduate Macroeconomics 2 Problem set 5. - Solutions

Machine Learning 4771

Fitting a transformation: Feature based alignment May 1 st, 2018

NPTEL Project. Econometric Modelling. Module23: Granger Causality Test. Lecture35: Granger Causality Test. Vinod Gupta School of Management

Speech recognition in noise by using word graph combinations

Water Hammer in Pipes

Lecture 28: Single Stage Frequency response. Context

Lecture 2 M/G/1 queues. M/G/1-queue

Reading. Lecture 28: Single Stage Frequency response. Lecture Outline. Context

Elastic Collisions. Definition: two point masses on which no external forces act collide without losing any energy.

INTRODUCTION TO MACHINE LEARNING 3RD EDITION

CHAPTER 10: LINEAR DISCRIMINATION

Robust and Accurate Cancer Classification with Gene Expression Profiling

Part II CONTINUOUS TIME STOCHASTIC PROCESSES

Combining Fuzzy Partitions Using Fuzzy Majority Vote and KNN

UNIVERSITAT AUTÒNOMA DE BARCELONA MARCH 2017 EXAMINATION

CHAPTER II AC POWER CALCULATIONS

[ ] 2. [ ]3 + (Δx i + Δx i 1 ) / 2. Δx i-1 Δx i Δx i+1. TPG4160 Reservoir Simulation 2018 Lecture note 3. page 1 of 5

Pattern Classification (III) & Pattern Verification

Pattern Classification (VI) 杜俊

Linear Response Theory: The connection between QFT and experiments

3D Human Pose Estimation from a Monocular Image Using Model Fitting in Eigenspaces

Hidden Markov Models

Time-line Hidden Markov Experts and its Application in Time Series Prediction

DEEP UNFOLDING FOR MULTICHANNEL SOURCE SEPARATION SUPPLEMENTARY MATERIAL

Physics 240: Worksheet 16 Name

Variants of Pegasos. December 11, 2009

P a g e 3 6 of R e p o r t P B 4 / 0 9

EEM 486: Computer Architecture

Dishonest casino as an HMM

Chapter 7 AC Power and Three-Phase Circuits

Chapter 4. Neural Networks Based on Competition

Physics 3 (PHYF144) Chap 3: The Kinetic Theory of Gases - 1

(,,, ) (,,, ). In addition, there are three other consumers, -2, -1, and 0. Consumer -2 has the utility function

Homework 8: Rigid Body Dynamics Due Friday April 21, 2017

Image Classification Using EM And JE algorithms

Lecture 12: HEMT AC Properties

MARKOV CHAIN AND HIDDEN MARKOV MODEL

Anomaly Detection. Lecture Notes for Chapter 9. Introduction to Data Mining, 2 nd Edition by Tan, Steinbach, Karpatne, Kumar

From Particles to Rigid Bodies

Homework 2 Solutions

y new = M x old Feature Selection: Linear Transformations Constraint Optimization (insertion)

A Cell Decomposition Approach to Online Evasive Path Planning and the Video Game Ms. Pac-Man

The Characterization of Jones Polynomial. for Some Knots

Bag for Sophia by Leonie Bateman and Deirdre Bond-Abel

Lecture 2 L n i e n a e r a M od o e d l e s

Comparison of several variants of the response spectrum method and definition of equivalent static loads from the peak response envelopes

Chapter 8. Linear Momentum, Impulse, and Collisions

Appendix to Online Clustering with Experts

グラフィカルモデルによる推論 確率伝搬法 (2) Kenji Fukumizu The Institute of Statistical Mathematics 計算推論科学概論 II (2010 年度, 後期 )

System in Weibull Distribution

TSS = SST + SSE An orthogonal partition of the total SS

On the number of regions in an m-dimensional space cut by n hyperplanes

Notes on the stability of dynamic systems and the use of Eigen Values.

CptS 570 Machine Learning School of EECS Washington State University. CptS Machine Learning 1

Video-Based Face Recognition Using Adaptive Hidden Markov Models

Transcription:

ETHEM ALPAYDIN The MIT Press, 2014 Lecure Sdes for INTRODUCTION TO MACHINE LEARNING 3RD EDITION aaydn@boun.edu.r h://www.ce.boun.edu.r/~ehe/23e

CHAPTER 7: CLUSTERING

Searaerc Densy Esaon 3 Paraerc: Assue a snge ode for ( C ) (Chaers 4 and 5) Searaerc: ( C ) s a ure of denses Mue ossbe eanaons/rooyes: Dfferen handwrng syes, accens n seech Nonaraerc: No ode; daa seaks for sef (Chaer 8)

Mure Denses 4 k G P G 1 where G he coonens/grous/cusers, P ( G ) ure roorons (rors), ( G ) coonen denses Gaussan ure where ( G ) ~ N ( μ, ) araeers Φ = {P ( G ), μ, } k =1 unabeed sae X={ } (unsuervsed earnng)

Casses vs. Cusers Suervsed: X = {,r } Casses C =1,...,K where ( C ) ~ N(μ, ) Φ = {P (C ), μ, } K =1 Unsuervsed : X = { } Cusers G =1,...,k where ( G ) ~ N ( μ, ) Φ = {P ( G ), μ, } k =1 Labes r? 5 k G P G 1 K P 1 C C T r r r r N r C P S ˆ

Fnd k reference vecors (rooyes/codebook vecors/codewords) whch bes reresen daa Reference vecors,, =1,...,k Use neares (os sar) reference: Reconsrucon error k-means Cuserng 6 n oherwse n f 0 1 1 k b b E X

Encodng/Decodng 7 b 1 f n 0 oherwse

8 k-eans Cuserng

9

Eecaon-Mazaon (EM) 10 Log kehood wh a ure ode L X og og k G P G 1 Assue hdden varabes z, whch when known, ake ozaon uch ser Coee kehood, L c (Φ X,Z), n ers of and z Incoee kehood, L(Φ X), n ers of

E- and M-ses 11 Ierae he wo ses 1. E-se: Esae z gven X and curren Φ 2. M-se: Fnd new Φ gven z, X, and od Φ. E - se: Q M- se: EL C X,Z 1 argaq X, An ncrease n Q ncreases ncoee kehood 1 X L X L

z = 1 f beongs o G, 0 oherwse (abes r of suervsed earnng); assue ( G )~N(μ, ) E-se: M-se: EM n Gaussan Mures 12 h G P G P G G P G z E,,, X, T h h h h N h P 1 1 1 1 S G Use esaed abes n ace of unknown abes

13 P(G 1 )=h 1 =0.5

Mures of Laen Varabe Modes 14 Reguarze cusers 1. Assue shared/dagona covarance arces 2. Use PCA/FA o decrease densonay: Mures of PCA/FA T N,V V ψ G Can use EM o earn V (Ghahraan and Hnon, 1997; Tng and Bsho, 1999)

Afer Cuserng 15 Densonay reducon ehods fnd correaons beween feaures and grou feaures Cuserng ehods fnd sares beween nsances and grou nsances Aows knowedge eracon hrough nuber of cusers, ror robabes, cuser araeers,.e., cener, range of feaures. Eae: CRM, cusoer segenaon

Cuserng as Prerocessng 16 Esaed grou abes h (sof) or b (hard) ay be seen as he densons of a new k densona sace, where we can hen earn our dscrnan or regressor. Loca reresenaon (ony one b s 1, a ohers are 0; ony few h are nonzero) vs Dsrbued reresenaon (Afer PCA; a z are nonzero)

Mure of Mures 17 In cassfcaon, he nu coes fro a ure of casses (suervsed). If each cass s aso a ure, e.g., of Gaussans, (unsuervsed), we have a ure of ures: k C G P G 1 K C PC 1

Secra Cuserng 18 Cuser usng redefned arwse sares B rs nsead of usng Eucdean or Mahaanobs dsance Can be used even f nsances no vecoray reresened Ses: I. Use Laacan Egenas (chaer 6) o a o a new z sace usng B rs II. Use k-eans n hs new z sace for cuserng

Herarchca Cuserng 19 Cuser based on sares/dsances Dsance easure beween nsances r and s Mnkowsk (L ) (Eucdean for = 2) d Cy-bock dsance r s d r s, 1 1/ d cb r s d r, 1 s

Aggoerave Cuserng 20 Sar wh N grous each wh one nsance and erge wo coses grous a each eraon Dsance beween wo grous G and G : Snge-nk: d r s G, G n d, r s G, G Coee-nk: d r s G, G a d, r s G, G Average-nk, cenrod d r s G, G ave d, r s G, G

Eae: Snge-Lnk Cuserng 21 Dendrogra

Choosng k 22 Defned by he acaon, e.g., age quanzaon Po daa (afer PCA) and check for cusers Increena (eader-cuser) agorh: Add one a a e un ebow (reconsrucon error/og kehood/nergrou dsances) Manuay check for eanng