Probabilistic Graphical Models

Similar documents
Exact Inference. Kayhan Batmanghelich

Machine Learning. Inference and Learning in GM. Eric Xing , Fall Lecture 18, November 10, b r a c e

Machine Learning. Graphical Models and Exact Inference. Eric Xing , Fall Lecture 17, November 7, Receptor A X 2.

Machine Learning. Recap of Basic Prob. Concepts

Why the Junction Tree Algorithm? The Junction Tree Algorithm. Clique Potential Representation. Overview. Chris Williams 1.

learning objectives learn what graphs are in mathematical terms learn how to represent graphs in computers learn about typical graph algorithms

, each of which is a tree, and whose roots r 1. , respectively, are children of r. Data Structures & File Management

Tangram Fractions Overview: Students will analyze standard and nonstandard

Weighted Graphs. Weighted graphs may be either directed or undirected.

Problem 1. Solution: = show that for a constant number of particles: c and V. a) Using the definitions of P

CSE 373: More on graphs; DFS and BFS. Michael Lee Wednesday, Feb 14, 2018

An undirected graph G = (V, E) V a set of vertices E a set of unordered edges (v,w) where v, w in V

OpenMx Matrices and Operators

Grade 7/8 Math Circles March 4/5, Graph Theory I- Solutions

Paths. Connectivity. Euler and Hamilton Paths. Planar graphs.

Cycles and Simple Cycles. Paths and Simple Paths. Trees. Problem: There is No Completely Standard Terminology!

Having a glimpse of some of the possibilities for solutions of linear systems, we move to methods of finding these solutions. The basic idea we shall

Outline. 1 Introduction. 2 Min-Cost Spanning Trees. 4 Example

12/3/12. Outline. Part 10. Graphs. Circuits. Euler paths/circuits. Euler s bridge problem (Bridges of Konigsberg Problem)

5/9/13. Part 10. Graphs. Outline. Circuits. Introduction Terminology Implementing Graphs

d e c b a d c b a d e c b a a c a d c c e b

Constructive Geometric Constraint Solving

CSE 373: AVL trees. Warmup: Warmup. Interlude: Exploring the balance invariant. AVL Trees: Invariants. AVL tree invariants review

ECE COMBINATIONAL BUILDING BLOCKS - INVEST 13 DECODERS AND ENCODERS

Outline. Computer Science 331. Computation of Min-Cost Spanning Trees. Costs of Spanning Trees in Weighted Graphs

16.unified Introduction to Computers and Programming. SOLUTIONS to Examination 4/30/04 9:05am - 10:00am

Problem solving by search

5/1/2018. Huffman Coding Trees. Huffman Coding Trees. Huffman Coding Trees. Huffman Coding Trees. Huffman Coding Trees. Huffman Coding Trees

S i m p l i f y i n g A l g e b r a SIMPLIFYING ALGEBRA.

Algorithmic and NP-Completeness Aspects of a Total Lict Domination Number of a Graph

V={A,B,C,D,E} E={ (A,D),(A,E),(B,D), (B,E),(C,D),(C,E)}

Exam 1 Solution. CS 542 Advanced Data Structures and Algorithms 2/14/2013

Math 166 Week in Review 2 Sections 1.1b, 1.2, 1.3, & 1.4

Theorem 1. An undirected graph is a tree if and only if there is a unique simple path between any two of its vertices.

CSE 373. Graphs 1: Concepts, Depth/Breadth-First Search reading: Weiss Ch. 9. slides created by Marty Stepp

1 Introduction to Modulo 7 Arithmetic

MAT3707. Tutorial letter 201/1/2017 DISCRETE MATHEMATICS: COMBINATORICS. Semester 1. Department of Mathematical Sciences MAT3707/201/1/2017

# 1 ' 10 ' 100. Decimal point = 4 hundred. = 6 tens (or sixty) = 5 ones (or five) = 2 tenths. = 7 hundredths.

V={A,B,C,D,E} E={ (A,D),(A,E),(B,D), (B,E),(C,D),(C,E)}

Applications of trees

In which direction do compass needles always align? Why?

Graphs Depth First Search

BASIC CAGE DETAILS SHOWN 3D MODEL: PSM ASY INNER WALL TABS ARE COINED OVER BASE AND COVER FOR RIGIDITY SPRING FINGERS CLOSED TOP

Bayesian belief networks

Strongly connected components. Finding strongly-connected components

CS 461, Lecture 17. Today s Outline. Example Run

Using the Printable Sticker Function. Using the Edit Screen. Computer. Tablet. ScanNCutCanvas

Outline. Circuits. Euler paths/circuits 4/25/12. Part 10. Graphs. Euler s bridge problem (Bridges of Konigsberg Problem)

CS 103 BFS Alorithm. Mark Redekopp

b. How many ternary words of length 23 with eight 0 s, nine 1 s and six 2 s?

CS61B Lecture #33. Administrivia: Autograder will run this evening. Today s Readings: Graph Structures: DSIJ, Chapter 12

Announcements. Not graphs. These are Graphs. Applications of Graphs. Graph Definitions. Graphs & Graph Algorithms. A6 released today: Risk

EE1000 Project 4 Digital Volt Meter

CMPS 2200 Fall Graphs. Carola Wenk. Slides courtesy of Charles Leiserson with changes and additions by Carola Wenk

Lecture 20: Minimum Spanning Trees (CLRS 23)

The University of Sydney MATH 2009

Module graph.py. 1 Introduction. 2 Graph basics. 3 Module graph.py. 3.1 Objects. CS 231 Naomi Nishimura

BASIC CAGE DETAILS D C SHOWN CLOSED TOP SPRING FINGERS INNER WALL TABS ARE COINED OVER BASE AND COVER FOR RIGIDITY

0.1. Exercise 1: the distances between four points in a graph

Functions and Graphs 1. (a) (b) (c) (f) (e) (d) 2. (a) (b) (c) (d)

CS 241 Analysis of Algorithms

CS200: Graphs. Graphs. Directed Graphs. Graphs/Networks Around Us. What can this represent? Sometimes we want to represent directionality:

PH427/PH527: Periodic systems Spring Overview of the PH427 website (syllabus, assignments etc.) 2. Coupled oscillations.

Garnir Polynomial and their Properties

Solutions for HW11. Exercise 34. (a) Use the recurrence relation t(g) = t(g e) + t(g/e) to count the number of spanning trees of v 1

The University of Sydney MATH2969/2069. Graph Theory Tutorial 5 (Week 12) Solutions 2008

Improving Union. Implementation. Union-by-size Code. Union-by-Size Find Analysis. Path Compression! Improving Find find(e)

CSC Design and Analysis of Algorithms. Example: Change-Making Problem

Graph Search (6A) Young Won Lim 5/18/18

Planar convex hulls (I)

4.1 Interval Scheduling. Chapter 4. Greedy Algorithms. Interval Scheduling: Greedy Algorithms. Interval Scheduling. Interval scheduling.

Module 2 Motion Instructions

Math 61 : Discrete Structures Final Exam Instructor: Ciprian Manolescu. You have 180 minutes.

SAMPLE CSc 340 EXAM QUESTIONS WITH SOLUTIONS: part 2

Depth First Search. Yufei Tao. Department of Computer Science and Engineering Chinese University of Hong Kong

Priority Search Trees - Part I

CS September 2018

QUESTIONS BEGIN HERE!

Multipoint Alternate Marking method for passive and hybrid performance monitoring

Graphs. Graphs. Graphs: Basic Terminology. Directed Graphs. Dr Papalaskari 1

Complete Solutions for MATH 3012 Quiz 2, October 25, 2011, WTT

Planar Upward Drawings

MULTIPLE-LEVEL LOGIC OPTIMIZATION II

(2) If we multiplied a row of B by λ, then the value is also multiplied by λ(here lambda could be 0). namely

Trees as operads. Lecture A formalism of trees

COMP108 Algorithmic Foundations

Edge-Triggered D Flip-flop. Formal Analysis. Fundamental-Mode Sequential Circuits. D latch: How do flip-flops work?

Experiment # 3 Introduction to Digital Logic Simulation and Xilinx Schematic Editor

Properties of Hexagonal Tile local and XYZ-local Series

More Foundations. Undirected Graphs. Degree. A Theorem. Graphs, Products, & Relations

Seven-Segment Display Driver

COMPLEXITY OF COUNTING PLANAR TILINGS BY TWO BARS

Register Allocation. How to assign variables to finitely many registers? What to do when it can t be done? How to do so efficiently?

GREEDY TECHNIQUE. Greedy method vs. Dynamic programming method:

Graphs. CSC 1300 Discrete Structures Villanova University. Villanova CSC Dr Papalaskari

CSE303 - Introduction to the Theory of Computing Sample Solutions for Exercises on Finite Automata

Graph Isomorphism. Graphs - II. Cayley s Formula. Planar Graphs. Outline. Is K 5 planar? The number of labeled trees on n nodes is n n-2

NP-Completeness. CS3230 (Algorithm) Traveling Salesperson Problem. What s the Big Deal? Given a Problem. What s the Big Deal? What s the Big Deal?

L.3922 M.C. L.3922 M.C. L.2996 M.C. L.3909 M.C. L.5632 M.C. L M.C. L.5632 M.C. L M.C. DRIVE STAR NORTH STAR NORTH NORTH DRIVE

Platform Controls. 1-1 Joystick Controllers. Boom Up/Down Controller Adjustments

Transcription:

Sool o oputr Sin roilisti rpil Mols xt Inrn: Vril liintion g H ri Xing Ltur 4 Jnury 27 2014 Ring: K-p 9 ri Xing @ MU 2005-2014 1

Rp: n: is prt p -p or istriution i II. ri Xing @ MU 2005-2014 2

Qustion: Is tr N tt is prt p or givn MN? T "ion" MN ri Xing @ MU 2005-2014 3

Qustion: Is tr N tt is prt p or givn MN? {} {} {} {} Tis MN os not v prt I-p s N! ri Xing @ MU 2005-2014 4

Qustion: Is tr n MN tt is prt I-p to givn N? V-strutur xpl ri Xing @ MU 2005-2014 5

Qustion: Is tr n MN tt is prt I-p to givn N? V-strutur s no quivlnt in MNs! ri Xing @ MU 2005-2014 6

rtilly irt yli rps lso ll in grps Nos n isjointly prtition into svrl in oponnts n g witin t s in oponnt ust unirt n g twn two nos in irnt in oponnts ust irt in oponnts: {} {} {}{}{H} {I} ri Xing @ MU 2005-2014 7

Sury Invstigt t rltionsip twn Ns n MNs Ty rprsnt irnt ilis o inpnn ssuptions Not ntion: in ntworks suprst o ot Ns n MNs Wy w r out tis: N n MN or irnt sntis or signr to ptur or xprssion onitionl inpnns ong vrils Unr rtin onition N n rprsnt s n MN n vi vrs In t utur or rtin oprtion i.. inrn w will using singl rprsnttion s t t strutur or wi n lgorit n oprt on. Tis ks lgorit sign n nlysis o t lgorits siplr ri Xing @ MU 2005-2014 8

roilisti Inrn n Lrning W now v opt rprsnttions o proility istriutions: rpil Mols M M sris uniqu proility istriution Typil tsks: Tsk 1: How o w nswr quris out M.g. M XY? W us inrn s n or t pross o oputing nswrs to su quris Tsk 2: How o w stit plusil ol M ro t? i. W us lrning s n or t pross o otining point stit o M. ii. ut or ysin ty sk pm wi is tully n inrn prol. iii. Wn not ll vrils r osrvl vn oputing point stit o M n to o inrn to iput t issing t. ri Xing @ MU 2005-2014 9

Qury 1: Liklioo Most o t quris on y sk involv vin vin is n ssignnt o vlus to st vrils in t oin Witout loss o gnrlity = { X k+1 X n } Siplst qury: oput proility o vin x x tis is otn rrr to s oputing t liklioo o x 1 x k 1 k ri Xing @ MU 2005-2014 10

Qury 2: onitionl roility Otn w r intrst in t onitionl proility istriution o vril givn t vin X X x X X x tis is t postriori li in X givn vin W usully qury sust Y o ll oin vrils X={YZ} n "on't r" out t rining Z: Y Y Z z t pross o suing out t "on't r" vrils z is ll rginliztion n t rsulting y is ll rginl pro. z ri Xing @ MU 2005-2014 11

pplitions o postriori li rition: wt is t proility o n outo givn t strting onition? t qury no is snnt o t vin ignosis: wt is t proility o iss/ult givn syptos t qury no n nstor o t vin Lrning unr prtil osrvtion? ill in t unosrv vlus unr n "M" stting or ltr T irtionlity o inortion low twn vrils is not rstrit y t irtionlity o t gs in M proilisti inrn n oin vin or ll prts o t ntwork ri Xing @ MU 2005-2014 12

xpl: p li Ntwork p li Ntwork N [Hinton t l. 2006] nrtiv ol wit ultipl in lyrs Sussul pplitions Rognizing nwrittn igits Lrning otion ptur t ollortiv iltring W 3 W 2 H 3 H 2 W 1 visil nos t H 1 V ri Xing @ MU 2005-2014 13

Qury 3: Most rol ssignnt In tis qury w wnt to in t ost prol joint ssignnt M or so vrils o intrst Su rsoning is usully pror unr so givn vin n ignoring t vlus o otr vrils z : M Y rg x yy y rg x yy z y z tis is t xiu postriori onigurtion o y. ri Xing @ MU 2005-2014 14

pplitions o M lssiition in ost likly ll givn t vin xplntion wt is t ost likly snrio givn t vin utionry not: T M o vril pns on its "ontxt"---t st o vrils n jointly quri xpl: M o Y 1? M o Y 1 Y 2? y 1 y 2 y 1 y 2 0 0 0.35 0 1 0.05 1 0 0.3 1 1 0.3 ri Xing @ MU 2005-2014 15

oplxity o Inrn T: oputing X = x in M is N-r Hrnss os not n w nnot solv inrn It iplis tt w nnot in gnrl prour tt works iintly or ritrry Ms or prtiulr ilis o Ms w n v provly iint prours ri Xing @ MU 2005-2014 16

ppros to inrn xt inrn lgorits T liintion lgorit Mssg-pssing lgorit su-prout li propgtion T juntion tr lgorits pproxit inrn tniqus Stosti siultion / spling tos Mrkov in Mont rlo tos Vritionl lgorits ri Xing @ MU 2005-2014 17

Mrginliztion n liintion signl trnsution ptwy: Wt is t liklioo tt protin is tiv? Qury: nïv sution ns to nurt ovr n xponntil nur o trs y in oposition w gt ri Xing @ MU 2005-2014 18

liintion on ins Rrrnging trs... ri Xing @ MU 2005-2014 19

Now w n pror innrost sution Tis sution "liints" on vril ro our sution rgunt t "lol ost". X p liintion on ins ri Xing @ MU 2005-2014 20

p p p X X liintion in ins Rrrnging n tn suing gin w gt ri Xing @ MU 2005-2014 21

liintion in ins X X X X liint nos on y on ll t wy to t n w gt oplxity: p stp osts OVlX i *VlX i+1 oprtions: Okn 2 opr to nïv vlution tt sus ovr joint vlus o n-1 vrils On k ri Xing @ MU 2005-2014 22

Hin Mrkov Mol y 1 y 2 y 3... y T px y = px 1 x T y 1 y T = py 1 px 1 y 1 py 2 y 1 px 2 y 2 py T y T-1 px T y T x 1 x 2 x 3... x T onitionl proility: ri Xing @ MU 2005-2014 23

Hin Mrkov Mol onitionl proility: y 1 x 1 y 2 y 3 x 2 x 3...... y T x T ri Xing @ MU 2005-2014 24

Rrrnging trs... Unirt ins Z Z 1 1 ri Xing @ MU 2005-2014 25

onitionl Rno ils Y 1 Y 2 Y 5 X 1 X n ri Xing @ MU 2005-2014 26

T Su-rout Oprtion In gnrl w n viw t tsk t n s tt o oputing t vlu o n xprssion o t or: z wr is st o tors W ll tis tsk t su-prout inrn tsk. ri Xing @ MU 2005-2014 27

Inrn on nrl M vi Vril liintion nrl i: Writ qury in t or tis suggsts n "liintion orr" o ltnt vrils to rginliz Itrtivly Mov ll irrlvnt trs outsi o innrost su ror innrost su gtting nw tr Insrt t nw tr into t prout wrp-up X 1 X 1 x n x x i 3 2 X 1 X x 1 x i p i 1 ri Xing @ MU 2005-2014 28

Outo o liintion Lt X so st o vrils lt st o tors su tt or Sop[ ] X lt Y X st o qury vrils n lt Z = X Y t vril to liint T rsult o liinting t vril Z is tor Y z Tis tor os not nssrily orrspon to ny proility or onitionl proility in tis ntwork. xpl ortoing ri Xing @ MU 2005-2014 29

ling wit vin onitioning s Su-rout Oprtion T vin potntil: Totl vin potntil: Introuing vin --- rstrit tors: i i i i i i i 0 i 1 z Y I i i i ri Xing @ MU 2005-2014 30

T liintion lgorit rour liintion // t M // vin Z // St o vrils to liint X // qury vrils 1. Initiliz 2. vin 3. Su-rout-liintion Z 4. Norliztion ri Xing @ MU 2005-2014 31

T liintion lgorit rour Initiliz Z 1. Lt Z 1... Z k n orring o Z su tt Z i Z j i i < j 2. Initiliz wit t ull t st o tors rour vin 1. or i = i i rour Su-rout-Vril- liintion Z 1. or i = 1... k Su-rout-liint-Vr Z i 2. 3. rturn 4. Norliztion ri Xing @ MU 2005-2014 32

T liintion lgorit rour Initiliz Z 1. Lt Z 1... Z k n orring o Z su tt Z i Z j i i < j 2. Initiliz wit t ull t st o tors rour vin 1. or i = i i rour Su-rout-Vril- liintion Z 1. or i = 1... k Su-rout-liint-Vr Z i 2. 3. rturn 4. Norliztion rour Norliztion 1. X= X/ x X rour Su-rout-liint-Vr // St o tors Z // Vril to liint 1. { : Z Sop[]} 2. Z 5. rturn {} ri Xing @ MU 2005-2014 33

or oplx ntwork oo w H Wt is t proility tt wks r lving givn tt t grss onition is poor? ri Xing @ MU 2005-2014 34

Qury: N to liint: H Initil tors: oos n liintion orr: H Stp 1: onitioning ix t vin no i.. on its osrv vlu i.. : Tis stp is isoorpi to rginliztion stp: H g ~ p ~ p ~ xpl: Vril liintion ri Xing @ MU 2005-2014 35

Qury: N to liint: Initil tors: Stp 2: liint oput H g g 1 g g g p g xpl: Vril liintion ri Xing @ MU 2005-2014 36

Qury: N to liint: Initil tors: Stp 3: liint oput H xpl: Vril liintion g g p ri Xing @ MU 2005-2014 37

Qury: N to liint: Initil tors: Stp 4: liint oput H xpl: Vril liintion g g p ri Xing @ MU 2005-2014 38

Qury: N to liint: Initil tors: Stp 5: liint oput H xpl: Vril liintion g g p ri Xing @ MU 2005-2014 39

Qury: N to liint: Initil tors: Stp 6: liint oput H xpl: Vril liintion p g g ri Xing @ MU 2005-2014 40

Qury: N to liint: Initil tors: Stp 7: liint oput H xpl: Vril liintion g g p ri Xing @ MU 2005-2014 41

Qury: N to liint: Initil tors: Stp 8: Wrp-up H xpl: Vril liintion g g ~ p p p p ~ p p ~ ri Xing @ MU 2005-2014 42

oplxity o vril liintion Suppos in on liintion stp w oput Tis rquirs x y1 yk ' x x y1 yk k Vl X Vl Y ultiplitions i or vlu or x y 1 y k w o k ultiplitions i ' x x Vl X Vl Y itions i y x k 1 yk i x y i i1 i or vlu o y 1 y k w o VlX itions oplxity is xponntil in nur o vrils in t intrit tor ri Xing @ MU 2005-2014 43

grp liintion lgorit orliztion H H grp liintion Unrstning Vril liintion ri Xing @ MU 2005-2014 44

rp liintion gin wit t unirt M or orliz N rp V n liintion orring I liint nxt no in t orring I Roving t no ro t grp onnting t rining nigors o t nos T ronstitut grp 'V ' Rtin t gs tt wr rt uring t liintion prour T grp-torti proprty: t tors rsult uring vril liintion r ptur y roring t liintion liqu ri Xing @ MU 2005-2014 45

grp liintion lgorit Intrit trs orrspon to t liqus rsult ro liintion orliztion H H grp liintion Unrstning Vril liintion H ri Xing @ MU 2005-2014 46

liintion liqus H H H g ri Xing @ MU 2005-2014 47

rp liintion n rginliztion Inu pnny uring rginliztion vs. liintion liqu Sution <-> liintion Intrit tr <-> liintion liqu H g g ri Xing @ MU 2005-2014 48

liqu tr H g g p ri Xing @ MU 2005-2014 49

oplxity T ovrll oplxity is trin y t nur o t lrgst liintion liqu Wt is t lrgst liintion liqu? pur grp torti qustion Tr-wit k: on lss tn t sllst ivl vlu o t rinlity o t lrgst liintion liqu rnging ovr ll possil liintion orring goo liintion orrings l to sll liqus n n ru oplxity wt will ppn i w liint "" irst in t ov grp? in t st liintion orring o grp --- N-r Inrn is N-r ut tr otn xist "ovious" optil or nr-opt liintion orring ri Xing @ MU 2005-2014 50

xpls Str Tr ri Xing @ MU 2005-2014 51

Mor xpl: Ising ol ri Xing @ MU 2005-2014 52

Liittion o rour liintion Liittion H H ri Xing @ MU 2005-2014 53

Our lgorit so r nswrs only on qury.g. on on no o w n to o oplt liintion or vry su qury? liintion ssg pssing on liqu tr Mssgs n rus H g H H ro liintion to Mssg ssing g p ri Xing @ MU 2005-2014 54

ro liintion to Mssg ssing Our lgorit so r nswrs only on qury.g. on on no o w n to o oplt liintion or vry su qury? liintion ssg pssing on liqu tr notr qury... g Mssgs n r rus otrs n to roput ri Xing @ MU 2005-2014 55 H

Sury T sipl liint lgorit pturs t ky lgoriti Oprtion unrlying proilisti inrn: --- Tt o tking su ovr prout o potntil untions Wt n w sy out t ovrll oputtionl oplxity o t lgorit? In prtiulr ow n w ontrol t "siz" o t suns tt ppr in t squn o sution oprtion. T oputtionl oplxity o t liint lgorit n ru to purly grp-torti onsirtions. Tis grp intrprttion will lso provi ints out ow to sign iprov inrn lgorit tt ovro t liittion o liint. ri Xing @ MU 2005-2014 56