SlideShare a Scribd company logo
1 of 30
Many random walks
are faster than one
Noga Alon Tel Aviv University
Chen Avin Ben Gurion University
Michal Koucky Czech Academy of Sciences
Gady Kozma Weizmann Institute
Zvi Lotker Ben Gurion University
Mark R. Tuttle Intel
Random walks
• Random step:
– Move to an adjacent node chosen at random (and uniformly)
• Random walk:
– Take an infinite sequence of random steps
Why random
walks are
good
Many Applications
• Graph exploration
– Randomization avoids need to know topology
– Randomization rules when the graph is changing or
unknown
• Communication: devices send messages at random
– Exhibits locality, simplicity, low-overhead, robustness
– Becoming a popular approach to mobile devices
What is bad
about
Random
walk?
Latency is a problem
• There are many measures of latency:
– Hitting time: Expected time E(Hi,j) to visit a given node
– Cover time: Expected time E(C) to visit all nodes
– Mixing time: Expected time to reach the stationary distribution
• Best cases:
– Dense, highly connected graphs such as the complete
graph, expanders, hypercube.
– Optimal cover time: Θ(n∙log n).
• Worst cases
– Connectivity “decreases” and bottlenecks exist.
– Examples: the lollipop graph Θ(n3), the line Θ(n2).
Cover Time - known results
Our question
Can multiple walks reduce the latency?
Choose a node v in a graph.
Start k random walks from node v.
Speed up
We define the speed-up of k-random walks on a graph G.
Our Questions:
Can The speed be k?
Can k walks cover the graph k time faster than 1
walk?
Our Answer:
Many times yes, but not always.
Outline
• First some fun: Calculate speed-ups for simple graphs
– Clique (complete graph): linear speed-up
– Barbell: exponential speed-up
– Cycle: logarithmic speed-up
• Then some answers: When is linear speed-up possible?
– General formulations of our linear speed-up result
• In terms of the ratio cover-time/hitting-time
• Conclusions and open problems
Calculating speed-ups for
simple graphs
Let’s calculate E(C1) and E(Ck) for
the clique, the barbell, the cycle.
The clique
Clique hitting time: n
• A random walk starting at node A
– Chooses a random node each step
– Chooses node B with probability 1/n
– Expected waiting time until choosing B is n
• Hitting time from A to B is n
• Example {1,2,1,3,1,2,5,4}
A
B
What about
cover time?
Coupon Collectors problem
• Let n types of coupons and each at
each trile a coupon is picked at uniform
random and independing .
• Find the earliest time at which all n
objects have been picked at least
once.
• We need nlog(n) time.
Clique speed-up: k
• A k-walk chooses nodes k times faster
– 1 step of a k-walk chooses k nodes at random
– k steps of a 1-walk chooses k nodes at random
• Calculate expectations, then regroup terms:
• Example {1,2,1,3,1,2,5,4}={1,1,1,5},{2,3,2,4}
)()( 1
ktCPtCP k

)()()()()( 111 k
t
k
tt
CkEtCkPtkCkPtCPCE  
The barbell
Barbell cover time: n2
• The walk starts at O and moves to L or R: let’s say to L
• The walk must move back to O in order to cover R
• How long do we expect to wait for this L  O transition?
– From L, the walk moves to O with probability 1/(n+1)
• Expect to fail n times and move to BL instead of O
– From BL, the walk takes a long time to return to L
• Remember the hitting time in the clique is n
• Expect n steps to return to L from inside BL
O RL
BL BR
21
))(()()( nnnOLECE 
Barbell speed-up: 2k
• Start k=c log(n) walks on O (but let’s ignore ugly constants)
• Expect half to move to BL, half to BR: that’s log(n) in each
• Expect log(n) walks in BL and BR to stay there for n steps
– Remember hitting time for the clique is n
• Expect log(n) walks in BL and BR to cover them in n steps
– Remember k-walk cover time for the clique is n log(n)/k
• So expect log(n) walks to cover barbell in n steps, not n2
– Trust me: Proof must turn each “expect” into “with high probability”
– Rejoice with me: That’s a speed up of n=2log(n) = 2k
O
BL BR
RL
The cycle
0
1n
i
i-1i+1
Cycle cover time: n2
• Let Ei be expected time to reach 0 from i
– E0 = 0
– Ei = 1 + Ei+1/2 + Ei-1/2
– En = E1
• Solve these recurrence relations
– Show Ei = (i-1)E1 - (i-1)i
• Notice Ei+1 – Ei = Ei – Ei-1 – 2
• Define Di+1 = Ei+1 – Ei and notice Di+1 = Di – 2 = E1 – 2i
– Show E1 ≈ n
• Notice E1 = En = (n-1)E1 - (n-1)n and solve for E1
• So Ei ≈ (i-1)n – (i-1)i = (i-1)(n-i)
– Maximized at i = n/2 and maximum value is n2/4
0
1n
i
i-1i+1
Cycle speed-up: log(k)
The probability for a single walk to cover the Cycle in n/2
steps is 2-n/2
.
So, 2Ω(n) walks are needed for a linear speed up.
Generalize for k, the speed up log(k)
0
1n
i
i-1i+1
Matthews’ Theorem
• Theorem: For any graph G
C1  H1 log (n)
• This bound may or may not be tight
– On a clique, the cover time is nlog(n) and hitting time is n
– On a line, the cover time and hitting time are both n2
When is Matthews’ Theorem
Tight
• Observations: Matthews is tight for many
important graphs:
– Cliques
– expanders,
– torus,
– hypercubes,
– d-dimensional grids,
– d-regular balanced trees,
– random graphs, etc.
-4
-2
0
2
4
-4
-2
0
2
4
-1
-0.5
0
0.5
1
-4
-2
0
2
4
linear speed-up
Matthews’ bound [MAT88]: for any graph G
Theorem: For any graph G and k ≤ log n
Corollary: linear speed up when Matthews’ bound is
tight.
Proof
(independent trails)
(Markov inequalit)
Main Results
Theorem: For a large collection of graphs, as long as k is not
too big there is a speed up of k-o(1).
The collection includes all graphs for which there is a gap
between the cover time C and hmax and k such:
24
Proof sketch: 2 speed-up
If there is a gap between the cover time C and hmax then
the cover time is concentrated.
Given: a single random walk of length C+o(C) covers the
graph with high probability.
We Show: 2 random walks of length C/2 + h covers the
graph w.h.p.
When h is an order less than C/2 we have a speed up of
2-o(1).
C/2+o(C)
C/2+o(C)
h
≥C+o(C)
v
u
Conclusion
• Observations: Matthews is tight for many
important graphs:
– Cliques
– expanders,
– torus,
– hypercubes,
– d-dimensional grids,
– d-regular balanced trees,
– random graphs, etc.
• We can prove a speed-up even when
Matthews is not tight …
-4
-2
0
2
4
-4
-2
0
2
4
-1
-0.5
0
0.5
1
-4
-2
0
2
4
The Result is Tight
For the 2-dimensional grid (torus)
– gives
– gives
On the grid could, ?
On the cycle could, ?
Fore a more restricted families of graphs we can have linear
speed up for a larger k.
Expanders
• Expanders yield impressive cover time speed-ups:
– We proved linear speed-up for many graphs for k  log n
– We can prove linear speed-up for expanders for k  n
Conclusions
• Linear speed-ups are possible for many important graphs
– Speed-ups are related to the ratio C1/H1 of cover and hitting times
– Linear speed-ups occur when this ratio is large
• Open problems:
– Is the speed-up always at most k? always at least log k?
– What is random walks can communicate or leave “breadcrumbs”?
– What if the graph is actually changing dynamically?
End

More Related Content

What's hot

Application of Integration
Application of IntegrationApplication of Integration
Application of Integrationtutorcircle1
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integrationtutorcircle1
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integrationtutorcircle1
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integrationtutorcircle1
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integrationtutorcircle1
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integrationtutorcircle1
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integrationtutorcircle1
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integrationtutorcircle1
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integrationtutorcircle1
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integrationtutorcircle1
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integrationtutorcircle1
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integrationtutorcircle1
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integrationtutorcircle1
 
Linear Transformations, Matrix Algebra
Linear Transformations, Matrix AlgebraLinear Transformations, Matrix Algebra
Linear Transformations, Matrix AlgebraPrasanth George
 
Bellman ford algorithm
Bellman ford algorithmBellman ford algorithm
Bellman ford algorithmRuchika Sinha
 
Free video lecture in india
Free video lecture in indiaFree video lecture in india
Free video lecture in indiaCss Founder
 
free Video lecture in india
free Video lecture in indiafree Video lecture in india
free Video lecture in indiaEdhole.com
 
Automata theory - Push Down Automata (PDA)
Automata theory - Push Down Automata (PDA)Automata theory - Push Down Automata (PDA)
Automata theory - Push Down Automata (PDA)Akila Krishnamoorthy
 

What's hot (20)

Application of Integration
Application of IntegrationApplication of Integration
Application of Integration
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integration
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integration
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integration
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integration
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integration
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integration
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integration
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integration
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integration
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integration
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integration
 
Application of Integration
Application of IntegrationApplication of Integration
Application of Integration
 
Linear Transformations, Matrix Algebra
Linear Transformations, Matrix AlgebraLinear Transformations, Matrix Algebra
Linear Transformations, Matrix Algebra
 
Bellman ford algorithm
Bellman ford algorithmBellman ford algorithm
Bellman ford algorithm
 
Free video lecture in india
Free video lecture in indiaFree video lecture in india
Free video lecture in india
 
Ece4510 notes03
Ece4510 notes03Ece4510 notes03
Ece4510 notes03
 
free Video lecture in india
free Video lecture in indiafree Video lecture in india
free Video lecture in india
 
Automata theory - Push Down Automata (PDA)
Automata theory - Push Down Automata (PDA)Automata theory - Push Down Automata (PDA)
Automata theory - Push Down Automata (PDA)
 
bode_plot By DEV
 bode_plot By DEV bode_plot By DEV
bode_plot By DEV
 

Similar to Zvi random-walks-slideshare

ICPC 2015, Tsukuba : Unofficial Commentary
ICPC 2015, Tsukuba: Unofficial CommentaryICPC 2015, Tsukuba: Unofficial Commentary
ICPC 2015, Tsukuba : Unofficial Commentaryirrrrr
 
Tower Of Hanoi -A MatheMatical PuZzle
Tower Of Hanoi -A MatheMatical PuZzleTower Of Hanoi -A MatheMatical PuZzle
Tower Of Hanoi -A MatheMatical PuZzlepurvanahar
 
Interpolating evolutionary tracks of rapidly rotating stars - presentation
Interpolating evolutionary tracks of rapidly rotating stars - presentationInterpolating evolutionary tracks of rapidly rotating stars - presentation
Interpolating evolutionary tracks of rapidly rotating stars - presentationDanielle Kumpulanian
 
Lecture 3 insertion sort and complexity analysis
Lecture 3   insertion sort and complexity analysisLecture 3   insertion sort and complexity analysis
Lecture 3 insertion sort and complexity analysisjayavignesh86
 
Bernard schutz gr
Bernard schutz grBernard schutz gr
Bernard schutz grjcklp1
 
Chapter One.pdf
Chapter One.pdfChapter One.pdf
Chapter One.pdfabay golla
 
Matt Purkeypile's Doctoral Dissertation Defense Slides
Matt Purkeypile's Doctoral Dissertation Defense SlidesMatt Purkeypile's Doctoral Dissertation Defense Slides
Matt Purkeypile's Doctoral Dissertation Defense Slidesmpurkeypile
 
1_Asymptotic_Notation_pptx.pptx
1_Asymptotic_Notation_pptx.pptx1_Asymptotic_Notation_pptx.pptx
1_Asymptotic_Notation_pptx.pptxpallavidhade2
 
HiPEAC'19 Tutorial on Quantum algorithms using QX - 2019-01-23
HiPEAC'19 Tutorial on Quantum algorithms using QX - 2019-01-23HiPEAC'19 Tutorial on Quantum algorithms using QX - 2019-01-23
HiPEAC'19 Tutorial on Quantum algorithms using QX - 2019-01-23Aritra Sarkar
 
A short introduction to Quantum Computing and Quantum Cryptography
A short introduction to Quantum Computing and Quantum CryptographyA short introduction to Quantum Computing and Quantum Cryptography
A short introduction to Quantum Computing and Quantum CryptographyFacultad de Informática UCM
 
Spread spectrum communications and CDMA
Spread spectrum communications and CDMASpread spectrum communications and CDMA
Spread spectrum communications and CDMAHossam Zein
 
Lect no 13 ECC.ppt
Lect no 13 ECC.pptLect no 13 ECC.ppt
Lect no 13 ECC.pptDEEPAK948083
 
Lect no 13 ECC.ppt
Lect no 13 ECC.pptLect no 13 ECC.ppt
Lect no 13 ECC.pptDEEPAK948083
 
Mit15 082 jf10_lec01
Mit15 082 jf10_lec01Mit15 082 jf10_lec01
Mit15 082 jf10_lec01Saad Liaqat
 
Black-Scholes overview
Black-Scholes overviewBlack-Scholes overview
Black-Scholes overviewInon Sharony
 

Similar to Zvi random-walks-slideshare (20)

ICPC 2015, Tsukuba : Unofficial Commentary
ICPC 2015, Tsukuba: Unofficial CommentaryICPC 2015, Tsukuba: Unofficial Commentary
ICPC 2015, Tsukuba : Unofficial Commentary
 
Lecture1
Lecture1Lecture1
Lecture1
 
Tower Of Hanoi -A MatheMatical PuZzle
Tower Of Hanoi -A MatheMatical PuZzleTower Of Hanoi -A MatheMatical PuZzle
Tower Of Hanoi -A MatheMatical PuZzle
 
Lecture24
Lecture24Lecture24
Lecture24
 
Interpolating evolutionary tracks of rapidly rotating stars - presentation
Interpolating evolutionary tracks of rapidly rotating stars - presentationInterpolating evolutionary tracks of rapidly rotating stars - presentation
Interpolating evolutionary tracks of rapidly rotating stars - presentation
 
Lecture 3 insertion sort and complexity analysis
Lecture 3   insertion sort and complexity analysisLecture 3   insertion sort and complexity analysis
Lecture 3 insertion sort and complexity analysis
 
Bernard schutz gr
Bernard schutz grBernard schutz gr
Bernard schutz gr
 
Ranging
RangingRanging
Ranging
 
8783733
87837338783733
8783733
 
Chapter One.pdf
Chapter One.pdfChapter One.pdf
Chapter One.pdf
 
Matt Purkeypile's Doctoral Dissertation Defense Slides
Matt Purkeypile's Doctoral Dissertation Defense SlidesMatt Purkeypile's Doctoral Dissertation Defense Slides
Matt Purkeypile's Doctoral Dissertation Defense Slides
 
1_Asymptotic_Notation_pptx.pptx
1_Asymptotic_Notation_pptx.pptx1_Asymptotic_Notation_pptx.pptx
1_Asymptotic_Notation_pptx.pptx
 
HiPEAC'19 Tutorial on Quantum algorithms using QX - 2019-01-23
HiPEAC'19 Tutorial on Quantum algorithms using QX - 2019-01-23HiPEAC'19 Tutorial on Quantum algorithms using QX - 2019-01-23
HiPEAC'19 Tutorial on Quantum algorithms using QX - 2019-01-23
 
A short introduction to Quantum Computing and Quantum Cryptography
A short introduction to Quantum Computing and Quantum CryptographyA short introduction to Quantum Computing and Quantum Cryptography
A short introduction to Quantum Computing and Quantum Cryptography
 
Teknik Simulasi
Teknik SimulasiTeknik Simulasi
Teknik Simulasi
 
Spread spectrum communications and CDMA
Spread spectrum communications and CDMASpread spectrum communications and CDMA
Spread spectrum communications and CDMA
 
Lect no 13 ECC.ppt
Lect no 13 ECC.pptLect no 13 ECC.ppt
Lect no 13 ECC.ppt
 
Lect no 13 ECC.ppt
Lect no 13 ECC.pptLect no 13 ECC.ppt
Lect no 13 ECC.ppt
 
Mit15 082 jf10_lec01
Mit15 082 jf10_lec01Mit15 082 jf10_lec01
Mit15 082 jf10_lec01
 
Black-Scholes overview
Black-Scholes overviewBlack-Scholes overview
Black-Scholes overview
 

More from Zvi Lotker

Variation on preferential-attachment
Variation  on  preferential-attachmentVariation  on  preferential-attachment
Variation on preferential-attachmentZvi Lotker
 
Asonam 2019-zvi-3-s
Asonam 2019-zvi-3-sAsonam 2019-zvi-3-s
Asonam 2019-zvi-3-sZvi Lotker
 
Knesset 17.07.2018 Zvi lotker talk on The mathematics of gender
Knesset 17.07.2018  Zvi lotker talk on The mathematics of genderKnesset 17.07.2018  Zvi lotker talk on The mathematics of gender
Knesset 17.07.2018 Zvi lotker talk on The mathematics of genderZvi Lotker
 
The effect of population control on societal fragmentation end-5
The effect of population control on societal fragmentation end-5The effect of population control on societal fragmentation end-5
The effect of population control on societal fragmentation end-5Zvi Lotker
 
Voting algorithm in the play julius 5
Voting algorithm in the play julius 5Voting algorithm in the play julius 5
Voting algorithm in the play julius 5Zvi Lotker
 
Small world effect
Small world effectSmall world effect
Small world effectZvi Lotker
 

More from Zvi Lotker (8)

Variation on preferential-attachment
Variation  on  preferential-attachmentVariation  on  preferential-attachment
Variation on preferential-attachment
 
Asonam 2019-zvi-3-s
Asonam 2019-zvi-3-sAsonam 2019-zvi-3-s
Asonam 2019-zvi-3-s
 
Knesset 17.07.2018 Zvi lotker talk on The mathematics of gender
Knesset 17.07.2018  Zvi lotker talk on The mathematics of genderKnesset 17.07.2018  Zvi lotker talk on The mathematics of gender
Knesset 17.07.2018 Zvi lotker talk on The mathematics of gender
 
Symmetry 2
Symmetry 2Symmetry 2
Symmetry 2
 
The effect of population control on societal fragmentation end-5
The effect of population control on societal fragmentation end-5The effect of population control on societal fragmentation end-5
The effect of population control on societal fragmentation end-5
 
Voting algorithm in the play julius 5
Voting algorithm in the play julius 5Voting algorithm in the play julius 5
Voting algorithm in the play julius 5
 
Leonard Cohen
Leonard CohenLeonard Cohen
Leonard Cohen
 
Small world effect
Small world effectSmall world effect
Small world effect
 

Recently uploaded

18-04-UA_REPORT_MEDIALITERAСY_INDEX-DM_23-1-final-eng.pdf
18-04-UA_REPORT_MEDIALITERAСY_INDEX-DM_23-1-final-eng.pdf18-04-UA_REPORT_MEDIALITERAСY_INDEX-DM_23-1-final-eng.pdf
18-04-UA_REPORT_MEDIALITERAСY_INDEX-DM_23-1-final-eng.pdfssuser54595a
 
Solving Puzzles Benefits Everyone (English).pptx
Solving Puzzles Benefits Everyone (English).pptxSolving Puzzles Benefits Everyone (English).pptx
Solving Puzzles Benefits Everyone (English).pptxOH TEIK BIN
 
Call Girls in Dwarka Mor Delhi Contact Us 9654467111
Call Girls in Dwarka Mor Delhi Contact Us 9654467111Call Girls in Dwarka Mor Delhi Contact Us 9654467111
Call Girls in Dwarka Mor Delhi Contact Us 9654467111Sapana Sha
 
Mastering the Unannounced Regulatory Inspection
Mastering the Unannounced Regulatory InspectionMastering the Unannounced Regulatory Inspection
Mastering the Unannounced Regulatory InspectionSafetyChain Software
 
SOCIAL AND HISTORICAL CONTEXT - LFTVD.pptx
SOCIAL AND HISTORICAL CONTEXT - LFTVD.pptxSOCIAL AND HISTORICAL CONTEXT - LFTVD.pptx
SOCIAL AND HISTORICAL CONTEXT - LFTVD.pptxiammrhaywood
 
Software Engineering Methodologies (overview)
Software Engineering Methodologies (overview)Software Engineering Methodologies (overview)
Software Engineering Methodologies (overview)eniolaolutunde
 
How to Configure Email Server in Odoo 17
How to Configure Email Server in Odoo 17How to Configure Email Server in Odoo 17
How to Configure Email Server in Odoo 17Celine George
 
Incoming and Outgoing Shipments in 1 STEP Using Odoo 17
Incoming and Outgoing Shipments in 1 STEP Using Odoo 17Incoming and Outgoing Shipments in 1 STEP Using Odoo 17
Incoming and Outgoing Shipments in 1 STEP Using Odoo 17Celine George
 
Organic Name Reactions for the students and aspirants of Chemistry12th.pptx
Organic Name Reactions  for the students and aspirants of Chemistry12th.pptxOrganic Name Reactions  for the students and aspirants of Chemistry12th.pptx
Organic Name Reactions for the students and aspirants of Chemistry12th.pptxVS Mahajan Coaching Centre
 
Introduction to ArtificiaI Intelligence in Higher Education
Introduction to ArtificiaI Intelligence in Higher EducationIntroduction to ArtificiaI Intelligence in Higher Education
Introduction to ArtificiaI Intelligence in Higher Educationpboyjonauth
 
“Oh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...
“Oh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...“Oh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...
“Oh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...Marc Dusseiller Dusjagr
 
Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...
Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...
Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...Krashi Coaching
 
MENTAL STATUS EXAMINATION format.docx
MENTAL     STATUS EXAMINATION format.docxMENTAL     STATUS EXAMINATION format.docx
MENTAL STATUS EXAMINATION format.docxPoojaSen20
 
KSHARA STURA .pptx---KSHARA KARMA THERAPY (CAUSTIC THERAPY)————IMP.OF KSHARA ...
KSHARA STURA .pptx---KSHARA KARMA THERAPY (CAUSTIC THERAPY)————IMP.OF KSHARA ...KSHARA STURA .pptx---KSHARA KARMA THERAPY (CAUSTIC THERAPY)————IMP.OF KSHARA ...
KSHARA STURA .pptx---KSHARA KARMA THERAPY (CAUSTIC THERAPY)————IMP.OF KSHARA ...M56BOOKSTORE PRODUCT/SERVICE
 
Concept of Vouching. B.Com(Hons) /B.Compdf
Concept of Vouching. B.Com(Hons) /B.CompdfConcept of Vouching. B.Com(Hons) /B.Compdf
Concept of Vouching. B.Com(Hons) /B.CompdfUmakantAnnand
 
Contemporary philippine arts from the regions_PPT_Module_12 [Autosaved] (1).pptx
Contemporary philippine arts from the regions_PPT_Module_12 [Autosaved] (1).pptxContemporary philippine arts from the regions_PPT_Module_12 [Autosaved] (1).pptx
Contemporary philippine arts from the regions_PPT_Module_12 [Autosaved] (1).pptxRoyAbrique
 

Recently uploaded (20)

18-04-UA_REPORT_MEDIALITERAСY_INDEX-DM_23-1-final-eng.pdf
18-04-UA_REPORT_MEDIALITERAСY_INDEX-DM_23-1-final-eng.pdf18-04-UA_REPORT_MEDIALITERAСY_INDEX-DM_23-1-final-eng.pdf
18-04-UA_REPORT_MEDIALITERAСY_INDEX-DM_23-1-final-eng.pdf
 
Solving Puzzles Benefits Everyone (English).pptx
Solving Puzzles Benefits Everyone (English).pptxSolving Puzzles Benefits Everyone (English).pptx
Solving Puzzles Benefits Everyone (English).pptx
 
Call Girls in Dwarka Mor Delhi Contact Us 9654467111
Call Girls in Dwarka Mor Delhi Contact Us 9654467111Call Girls in Dwarka Mor Delhi Contact Us 9654467111
Call Girls in Dwarka Mor Delhi Contact Us 9654467111
 
Mastering the Unannounced Regulatory Inspection
Mastering the Unannounced Regulatory InspectionMastering the Unannounced Regulatory Inspection
Mastering the Unannounced Regulatory Inspection
 
SOCIAL AND HISTORICAL CONTEXT - LFTVD.pptx
SOCIAL AND HISTORICAL CONTEXT - LFTVD.pptxSOCIAL AND HISTORICAL CONTEXT - LFTVD.pptx
SOCIAL AND HISTORICAL CONTEXT - LFTVD.pptx
 
Software Engineering Methodologies (overview)
Software Engineering Methodologies (overview)Software Engineering Methodologies (overview)
Software Engineering Methodologies (overview)
 
How to Configure Email Server in Odoo 17
How to Configure Email Server in Odoo 17How to Configure Email Server in Odoo 17
How to Configure Email Server in Odoo 17
 
TataKelola dan KamSiber Kecerdasan Buatan v022.pdf
TataKelola dan KamSiber Kecerdasan Buatan v022.pdfTataKelola dan KamSiber Kecerdasan Buatan v022.pdf
TataKelola dan KamSiber Kecerdasan Buatan v022.pdf
 
Incoming and Outgoing Shipments in 1 STEP Using Odoo 17
Incoming and Outgoing Shipments in 1 STEP Using Odoo 17Incoming and Outgoing Shipments in 1 STEP Using Odoo 17
Incoming and Outgoing Shipments in 1 STEP Using Odoo 17
 
9953330565 Low Rate Call Girls In Rohini Delhi NCR
9953330565 Low Rate Call Girls In Rohini  Delhi NCR9953330565 Low Rate Call Girls In Rohini  Delhi NCR
9953330565 Low Rate Call Girls In Rohini Delhi NCR
 
Organic Name Reactions for the students and aspirants of Chemistry12th.pptx
Organic Name Reactions  for the students and aspirants of Chemistry12th.pptxOrganic Name Reactions  for the students and aspirants of Chemistry12th.pptx
Organic Name Reactions for the students and aspirants of Chemistry12th.pptx
 
Introduction to ArtificiaI Intelligence in Higher Education
Introduction to ArtificiaI Intelligence in Higher EducationIntroduction to ArtificiaI Intelligence in Higher Education
Introduction to ArtificiaI Intelligence in Higher Education
 
“Oh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...
“Oh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...“Oh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...
“Oh GOSH! Reflecting on Hackteria's Collaborative Practices in a Global Do-It...
 
Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...
Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...
Kisan Call Centre - To harness potential of ICT in Agriculture by answer farm...
 
MENTAL STATUS EXAMINATION format.docx
MENTAL     STATUS EXAMINATION format.docxMENTAL     STATUS EXAMINATION format.docx
MENTAL STATUS EXAMINATION format.docx
 
Staff of Color (SOC) Retention Efforts DDSD
Staff of Color (SOC) Retention Efforts DDSDStaff of Color (SOC) Retention Efforts DDSD
Staff of Color (SOC) Retention Efforts DDSD
 
KSHARA STURA .pptx---KSHARA KARMA THERAPY (CAUSTIC THERAPY)————IMP.OF KSHARA ...
KSHARA STURA .pptx---KSHARA KARMA THERAPY (CAUSTIC THERAPY)————IMP.OF KSHARA ...KSHARA STURA .pptx---KSHARA KARMA THERAPY (CAUSTIC THERAPY)————IMP.OF KSHARA ...
KSHARA STURA .pptx---KSHARA KARMA THERAPY (CAUSTIC THERAPY)————IMP.OF KSHARA ...
 
Concept of Vouching. B.Com(Hons) /B.Compdf
Concept of Vouching. B.Com(Hons) /B.CompdfConcept of Vouching. B.Com(Hons) /B.Compdf
Concept of Vouching. B.Com(Hons) /B.Compdf
 
Código Creativo y Arte de Software | Unidad 1
Código Creativo y Arte de Software | Unidad 1Código Creativo y Arte de Software | Unidad 1
Código Creativo y Arte de Software | Unidad 1
 
Contemporary philippine arts from the regions_PPT_Module_12 [Autosaved] (1).pptx
Contemporary philippine arts from the regions_PPT_Module_12 [Autosaved] (1).pptxContemporary philippine arts from the regions_PPT_Module_12 [Autosaved] (1).pptx
Contemporary philippine arts from the regions_PPT_Module_12 [Autosaved] (1).pptx
 

Zvi random-walks-slideshare

  • 1. Many random walks are faster than one Noga Alon Tel Aviv University Chen Avin Ben Gurion University Michal Koucky Czech Academy of Sciences Gady Kozma Weizmann Institute Zvi Lotker Ben Gurion University Mark R. Tuttle Intel
  • 2. Random walks • Random step: – Move to an adjacent node chosen at random (and uniformly) • Random walk: – Take an infinite sequence of random steps Why random walks are good
  • 3. Many Applications • Graph exploration – Randomization avoids need to know topology – Randomization rules when the graph is changing or unknown • Communication: devices send messages at random – Exhibits locality, simplicity, low-overhead, robustness – Becoming a popular approach to mobile devices What is bad about Random walk?
  • 4. Latency is a problem • There are many measures of latency: – Hitting time: Expected time E(Hi,j) to visit a given node – Cover time: Expected time E(C) to visit all nodes – Mixing time: Expected time to reach the stationary distribution
  • 5. • Best cases: – Dense, highly connected graphs such as the complete graph, expanders, hypercube. – Optimal cover time: Θ(n∙log n). • Worst cases – Connectivity “decreases” and bottlenecks exist. – Examples: the lollipop graph Θ(n3), the line Θ(n2). Cover Time - known results
  • 6. Our question Can multiple walks reduce the latency? Choose a node v in a graph. Start k random walks from node v.
  • 7. Speed up We define the speed-up of k-random walks on a graph G. Our Questions: Can The speed be k? Can k walks cover the graph k time faster than 1 walk? Our Answer: Many times yes, but not always.
  • 8. Outline • First some fun: Calculate speed-ups for simple graphs – Clique (complete graph): linear speed-up – Barbell: exponential speed-up – Cycle: logarithmic speed-up • Then some answers: When is linear speed-up possible? – General formulations of our linear speed-up result • In terms of the ratio cover-time/hitting-time • Conclusions and open problems
  • 9. Calculating speed-ups for simple graphs Let’s calculate E(C1) and E(Ck) for the clique, the barbell, the cycle.
  • 11. Clique hitting time: n • A random walk starting at node A – Chooses a random node each step – Chooses node B with probability 1/n – Expected waiting time until choosing B is n • Hitting time from A to B is n • Example {1,2,1,3,1,2,5,4} A B What about cover time?
  • 12. Coupon Collectors problem • Let n types of coupons and each at each trile a coupon is picked at uniform random and independing . • Find the earliest time at which all n objects have been picked at least once. • We need nlog(n) time.
  • 13. Clique speed-up: k • A k-walk chooses nodes k times faster – 1 step of a k-walk chooses k nodes at random – k steps of a 1-walk chooses k nodes at random • Calculate expectations, then regroup terms: • Example {1,2,1,3,1,2,5,4}={1,1,1,5},{2,3,2,4} )()( 1 ktCPtCP k  )()()()()( 111 k t k tt CkEtCkPtkCkPtCPCE  
  • 15. Barbell cover time: n2 • The walk starts at O and moves to L or R: let’s say to L • The walk must move back to O in order to cover R • How long do we expect to wait for this L  O transition? – From L, the walk moves to O with probability 1/(n+1) • Expect to fail n times and move to BL instead of O – From BL, the walk takes a long time to return to L • Remember the hitting time in the clique is n • Expect n steps to return to L from inside BL O RL BL BR 21 ))(()()( nnnOLECE 
  • 16. Barbell speed-up: 2k • Start k=c log(n) walks on O (but let’s ignore ugly constants) • Expect half to move to BL, half to BR: that’s log(n) in each • Expect log(n) walks in BL and BR to stay there for n steps – Remember hitting time for the clique is n • Expect log(n) walks in BL and BR to cover them in n steps – Remember k-walk cover time for the clique is n log(n)/k • So expect log(n) walks to cover barbell in n steps, not n2 – Trust me: Proof must turn each “expect” into “with high probability” – Rejoice with me: That’s a speed up of n=2log(n) = 2k O BL BR RL
  • 18. Cycle cover time: n2 • Let Ei be expected time to reach 0 from i – E0 = 0 – Ei = 1 + Ei+1/2 + Ei-1/2 – En = E1 • Solve these recurrence relations – Show Ei = (i-1)E1 - (i-1)i • Notice Ei+1 – Ei = Ei – Ei-1 – 2 • Define Di+1 = Ei+1 – Ei and notice Di+1 = Di – 2 = E1 – 2i – Show E1 ≈ n • Notice E1 = En = (n-1)E1 - (n-1)n and solve for E1 • So Ei ≈ (i-1)n – (i-1)i = (i-1)(n-i) – Maximized at i = n/2 and maximum value is n2/4 0 1n i i-1i+1
  • 19. Cycle speed-up: log(k) The probability for a single walk to cover the Cycle in n/2 steps is 2-n/2 . So, 2Ω(n) walks are needed for a linear speed up. Generalize for k, the speed up log(k) 0 1n i i-1i+1
  • 20. Matthews’ Theorem • Theorem: For any graph G C1  H1 log (n) • This bound may or may not be tight – On a clique, the cover time is nlog(n) and hitting time is n – On a line, the cover time and hitting time are both n2
  • 21. When is Matthews’ Theorem Tight • Observations: Matthews is tight for many important graphs: – Cliques – expanders, – torus, – hypercubes, – d-dimensional grids, – d-regular balanced trees, – random graphs, etc. -4 -2 0 2 4 -4 -2 0 2 4 -1 -0.5 0 0.5 1 -4 -2 0 2 4
  • 22. linear speed-up Matthews’ bound [MAT88]: for any graph G Theorem: For any graph G and k ≤ log n Corollary: linear speed up when Matthews’ bound is tight.
  • 24. Main Results Theorem: For a large collection of graphs, as long as k is not too big there is a speed up of k-o(1). The collection includes all graphs for which there is a gap between the cover time C and hmax and k such: 24
  • 25. Proof sketch: 2 speed-up If there is a gap between the cover time C and hmax then the cover time is concentrated. Given: a single random walk of length C+o(C) covers the graph with high probability. We Show: 2 random walks of length C/2 + h covers the graph w.h.p. When h is an order less than C/2 we have a speed up of 2-o(1). C/2+o(C) C/2+o(C) h ≥C+o(C) v u
  • 26. Conclusion • Observations: Matthews is tight for many important graphs: – Cliques – expanders, – torus, – hypercubes, – d-dimensional grids, – d-regular balanced trees, – random graphs, etc. • We can prove a speed-up even when Matthews is not tight … -4 -2 0 2 4 -4 -2 0 2 4 -1 -0.5 0 0.5 1 -4 -2 0 2 4
  • 27. The Result is Tight For the 2-dimensional grid (torus) – gives – gives On the grid could, ? On the cycle could, ? Fore a more restricted families of graphs we can have linear speed up for a larger k.
  • 28. Expanders • Expanders yield impressive cover time speed-ups: – We proved linear speed-up for many graphs for k  log n – We can prove linear speed-up for expanders for k  n
  • 29. Conclusions • Linear speed-ups are possible for many important graphs – Speed-ups are related to the ratio C1/H1 of cover and hitting times – Linear speed-ups occur when this ratio is large • Open problems: – Is the speed-up always at most k? always at least log k? – What is random walks can communicate or leave “breadcrumbs”? – What if the graph is actually changing dynamically?
  • 30. End