on the size of memory : human memory modeling by simulation (hum2s)

35
On The Size of Memory On The Size of Memory : : Human Memory Modeling by Human Memory Modeling by Simulation (HuM2S) Simulation (HuM2S) Haluk Bing Haluk Bing o o l l Complex Systems Lab Complex Systems Lab Dept. of Computer Engineering Dept. of Computer Engineering Bogazici University Bogazici University Ulusal Grid Çalıştayı Ulusal Grid Çalıştayı 200 200 7 7 Mar Mar 01 01 , 200 , 200 7 7

Upload: ellard

Post on 14-Jan-2016

19 views

Category:

Documents


2 download

DESCRIPTION

On The Size of Memory : Human Memory Modeling by Simulation (HuM2S). Haluk Bing o l Complex Systems Lab Dept. of Computer Engineering Bogazici University Ulusal Grid Çalıştayı 200 7 Mar 01 , 200 7. Outline. Motivation Memory Representation Recommendation Simple Recommendation Model - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

On The Size of MemoryOn The Size of Memory::Human Memory Modeling by Human Memory Modeling by Simulation (HuM2S)Simulation (HuM2S)

Haluk BingHaluk BingoollComplex Systems LabComplex Systems LabDept. of Computer EngineeringDept. of Computer EngineeringBogazici UniversityBogazici University

Ulusal Grid ÇalıştayıUlusal Grid Çalıştayı 200 20077MarMar 0101, 200, 20077

Page 2: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

2/33

OutlineOutline• MotivationMotivation• Memory RepresentationMemory Representation• RecommendationRecommendation• Simple Recommendation ModelSimple Recommendation Model• Effect of Small Memory SizeEffect of Small Memory Size• ConclusionsConclusions

[Bingol, [Bingol, LNCS 3733 pp.294-303LNCS 3733 pp.294-303 (2005)] (2005)][Bingol, arXiv nlin.AO/0609033, 2006 ][Bingol, arXiv nlin.AO/0609033, 2006 ]

Page 3: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

3/33

MotivationMotivation

Page 4: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

4/33

Motivation Motivation

If you need somebodyIf you need somebody

If you need a plumber, how do you find If you need a plumber, how do you find one?one?

• know one?know one?– You ?You ?– Your friends ?Your friends ?– Google ?Google ?

• FamousFamous plumber! plumber!

Page 5: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

5/33

Motivation Motivation

Key conceptsKey concepts• My My memorymemory• My friends memoryMy friends memory

My My networknetwork of friends of friendsLocalLocal memory memory

• Google’s memoryGoogle’s memoryAccessibleAccessible memory of the memory of the populationpopulationGlobalGlobal memory memory

• FameFameBeing Being knownknown to the general public to the general public

• RecommendationRecommendation

Page 6: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

6/33

RepresentationRepresentation(Static View)(Static View)

Page 7: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

7/33

22

Memory Memory Representation Representation

Who-knows-who GraphWho-knows-who Graph• Digraph Digraph G G ((PP, , AA))• Vertices Vertices PP ={1, 2, ..., ={1, 2, ..., nn}}

• Arc from Arc from ppii to to ppjj – if if ppii knows knows ppjj

11

33

55

44

44 22 55 3311

44 3322

44

4455

--

44 11 5533

Page 8: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

8/33

• MemoryMemory of of ppii

– MMii = {= {ppjj | | ppii knows knows ppjj } ⊆ } ⊆ PP

• mm : memory size of an individual : memory size of an individual– ||MMi i || = = mm

• nn : population size : population size– mn mn : memory capacity of the : memory capacity of the populationpopulation

......1 2 ... 1 2 ... mm

......

......

......

: : ... :: : ... :

11

22

33

::

nn

11

22

mm

......

abstractionabstraction

......1 2 ... 1 2 ... mm

......

......

......

: : ... :: : ... :

11

22

33

::

nn

......1 2 ... 1 2 ... mm

......

......

......

: : ... :: : ... :

11

22

33

::

nn

Memory Memory Representation Representation

MemoryMemory

• RemarksRemarks– out-degree: out-degree: mm

Page 9: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

9/33

• Knownness Knownness – kki i == ||{{ppi i | | ppi i knows knows ppjj}}||

11

22

mm

...... ......

Memory Memory Representation Representation

DefinitionsDefinitions

......1 2 ... 1 2 ... mm

......

......

......

: : ... :: : ... :

11

22

33

::

nn

......1 2 ... 1 2 ... mm

......

......

......

: : ... :: : ... :

11

22

33

::

nn

......1 2 ... 1 2 ... mm

......

......

......

: : ... :: : ... :

11

22

33

::

nn

• RemarksRemarks– out-degree: out-degree: mm – in-degree: 0 ≤ in-degree: 0 ≤ kkii ≤ ≤ nn

Page 10: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

10/33

Interacting Agents:Interacting Agents: RecommendationRecommendation

->->

Page 11: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

11/33

RecommendationRecommendation

Recommendation ProcessRecommendation ProcessPerson (from)Person (from)

• RememberRemember• RecommendRecommend• LearnLearn

Person (to)Person (to)

timetime

Page 12: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

12/33

RecommendationRecommendation

fromfrom

• Select a person Select a person ppff

– Select Select ff in {1, 2, ... , in {1, 2, ... , nn}}

• ppf f remembersremembers a known person a known person– Select Select rr in {1, 2, ... , in {1, 2, ... , mm}}

• Find person Find person ppr r

......1 2 ... 1 2 ... mm

......

......

......: : ... :: : ... :

11

22

33

::

nn

pprrff

rr

• f f : : ffromrom• rr : : rrecommendedecommended

Page 13: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

13/33

......1 2 ... 1 2 ... mm

......

......

......: : ... :: : ... :

11

22

33

::

nn

RecommendationRecommendation

toto

• Select a person Select a person pptt

– Select Select tt in {1, 2, ... , in {1, 2, ... , nn}}

• Recommend Recommend ppr r to to pptt

ff pprr

rr

tt

• t t : : ttoo

Page 14: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

14/33

......1 2 ... 1 2 ... mm

......

......

......: : ... :: : ... :

11

22

33

::

nn

RecommendationRecommendation

toto

• LearnLearn recommended person recommended person pprr

– ppt t remembersremembers a known person a known person• Select Select ee in {1, 2, ... , in {1, 2, ... , mm}}

– Person Person ppee

ff pprr

rr

tt

ee

ppee

• e e : slot to be : slot to be eemptiedmptied

Page 15: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

15/33

......1 2 ... 1 2 ... mm

......

......

......: : ... :: : ... :

11

22

33

::

nn

RecommendationRecommendation

toto• LearnLearn recommended person recommended person pprr

– ppt t remembersremembers a known person a known person• Select Select ee in {1, 2, ... , in {1, 2, ... , mm}}

– Person Person ppee

– Forget a personForget a person• Obtain an empty memory locationObtain an empty memory location

– Put Put ppr r to locationto location t-bt-bff pprr

rr

tt

ee

pprr

Page 16: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

16/33

Simple Simple RecommendationRecommendation ModelModel

->->

Page 17: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

17/33

Simple Recommendation Simple Recommendation ModelModel• Random selections Random selections

– f, tf, t ∊∊ {1, 2, ... , {1, 2, ... , nn}}– r, r, ee ∊∊ {1, 2, ... , {1, 2, ... , mm} }

• Initial memory contentInitial memory content– Every person knows next Every person knows next mm personspersons

Page 18: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

18/33

Effect of Small Memory SizeEffect of Small Memory Size

Page 19: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

19/33

Effect of Small Memory SizeEffect of Small Memory Size

DefinitionsDefinitions• Memory ratioMemory ratio

– ρ ρ = = m m / / nn– 0 < ρ 0 < ρ ≪ 1≪ 1

• FameFame– ffii = = kki i //nn

PopulatioPopulationn

nn mm ρρ

NDCOSNDCOS 7070 55 7 x 107 x 10--

11

IstanbulIstanbul 15 M15 M 1 k1 k 6 x 106 x 10--

55

TurkeyTurkey 73 M73 M 2 k2 k 3 x 103 x 10--

55

Page 20: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

20/33

Effect of Small Memory SizeEffect of Small Memory Size

Effect of Effect of ρρAs ρAs ρ→0→0

• ffminmin ↘ ↘– Some completely Some completely forgottenforgotten

• ffmaxmax ↗ ↗– A few gets more A few gets more knownknown

nn = 100 = 100

ffminmin

ffmaxmax

Completely Completely forgottenforgotten

Page 21: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

21/33

Effect of Small Memory SizeEffect of Small Memory Size

Minimum Fame Minimum Fame As As ρρ→→00

• ffminmin ↘ ↘

• LinearLinear

Page 22: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

22/33

Effect of Small Memory SizeEffect of Small Memory Size

Completely Forgotten Completely Forgotten AgentsAgentsc c : the number of : the number of

completely completely fforgotten agents orgotten agents

u u = = cc / / nn

As As ρρ→→00• uu ↗↗• LinearLinear

Page 23: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

23/33

Effect of Small Memory SizeEffect of Small Memory Size

Maximum FameMaximum FameAs As ρρ→→00

• ffmmaxax ↘↗↘↗

• LinearLinear

• Emergence of fameEmergence of fame

Page 24: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

24/33

Effect of Small Memory SizeEffect of Small Memory Size

Top 5% FameTop 5% FameAs As ρρ→→00

• ff5%5% : : Top 5% fame Top 5% fame

• ff5%5% ↘↘

• LinearLinear

Page 25: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

25/33

ConclusionsConclusions

• A new modelA new model–““too little memory, too little memory, too many items” casestoo many items” cases–Information Information disseminationdissemination–Building consensusBuilding consensus–AdvertisementAdvertisement

• Linear patternLinear patternss–ffminmin ↘ ↘–uu ↗↗–ffmaxmax ↘↗ ↘↗

• Emergence of Emergence of “fame” “fame” • Population Population PP could could be anythingbe anything

–PapersPapers–BooksBooks–MoviesMovies–Movie StarsMovie Stars–PaintersPainters–PoetsPoets–......

http://www.cmpe.boun.edu.tr/soslabhttp://www.cmpe.boun.edu.tr/soslab

Page 26: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

26/33

Human Memory Modeling by Human Memory Modeling by Simulation (HuM2S) ProjectSimulation (HuM2S) Project

Page 27: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

27/33

Project DescriptionProject Description

• An agent-based Complex Systems An agent-based Complex Systems modeling and simulation projectmodeling and simulation project

• Human population is studiedHuman population is studied– Human memory is modeled Human memory is modeled – Interaction of humans are simulatedInteraction of humans are simulated– Effect of memory size is investigatedEffect of memory size is investigated

Page 28: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

28/33

Potential OutcomesPotential Outcomes

•Understanding Understanding dynamics of human dynamics of human populationpopulation

–FameFame–Cultural islandsCultural islands–Integration of culturesIntegration of cultures–Dissemination of Dissemination of informationinformation–Building consensusBuilding consensus–Building public opinionBuilding public opinion–CooperationCooperation–Effects of Effects of advertisementadvertisement

• A new A new agent-based agent-based modelmodel for complex for complex sytemssytems

–““too little memory, too too little memory, too many items” casesmany items” cases–SynchronizationSynchronization

Page 29: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

29/33

Need for GridNeed for Grid

• Currently a single simulation takes Currently a single simulation takes months on a single machinemonths on a single machine

• Parametric parallelization possibleParametric parallelization possible– Number of persons currently 1.000Number of persons currently 1.000

• need for 1.000.000need for 1.000.000

– Memory sizes of currently 200Memory sizes of currently 200• need 1.000need 1.000

– Number of interactions currently 10Number of interactions currently 1088 • need 10need 101010

Page 30: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

30/33

BudgetBudget

• Research Assistant Research Assistant – 12 man-months12 man-months

• Notebook ?Notebook ?– 2.500 Euro2.500 Euro

• Conference/Travel ?Conference/Travel ?– 2.000 Euro2.000 Euro

Page 31: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

31/33

HuM2S ProjectHuM2S Project

• Just startedJust started– Feb 1, 2007Feb 1, 2007

• 25 000 Euro25 000 Euro– AssistantAssistant

Page 32: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

32/33

ReferencesReferences

• Bingol, Fame as an Effect of the Memory Size, Bingol, Fame as an Effect of the Memory Size, LNCS LNCS 37333733,, 294 (2005)294 (2005)

• Bingol, Fame as an Effect of the Memory Size, Bingol, Fame as an Effect of the Memory Size, ECCS’05 (presented), (2005)ECCS’05 (presented), (2005)

• Bingol, Bingol, On The Size of MemoryOn The Size of Memory, NDCOS , NDCOS (presented), (2006)(presented), (2006)

• Bingol, Emergence of Fame, PRE (submitted)Bingol, Emergence of Fame, PRE (submitted)preprint arXiv nlin.AO/0609033, 2006 preprint arXiv nlin.AO/0609033, 2006

Page 33: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

33/33

Thank YouThank You

http://www.cmpe.boun.edu.tr/http://www.cmpe.boun.edu.tr/soslabsoslab

Page 34: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

34/33

Future workFuture work

• Larger populationLarger population and memory and memory– Currently Currently nn=1 000 & =1 000 & mm=900=900

• Interaction of societiesInteraction of societies– Population size vs memoryPopulation size vs memory

• Effect of advertisementEffect of advertisement– Recommendation prefers someRecommendation prefers some

Page 35: On The Size of Memory : Human Memory Modeling by Simulation (HuM2S)

Bin

gol

35/33

RepresentationRepresentation Memory ContentMemory Content

jj thth memory content of person memory content of person ppii

ppkk ∈∈MMii

......1 2 ... 1 2 ... mm

......

......

......

: : ... :: : ... :

11

22

33

::

nn

......1 2 ... 1 2 ... mm

......

......

......

: : ... :: : ... :

11

22

33

::

nn

......1 2 ... 1 2 ... mm

......

......

......

: : ... :: : ... :

11

22

33

::

nn

......ppkkii

jj