Emergent a syst em C-kur s, 5 pong, HT-04 J onny Pet t ersson j - - PDF document

1

2/11 - 04 1 Emergent Systems, Jonny Pettersson, UmU

Emergent a syst em

C-kur s, 5 poäng, HT-04

J onny Pet t ersson

j onny@cs.umu.se

2/11 - 04 2 Emergent Systems, Jonny Pettersson, UmU

Cource Descript ion

❒ The f ocus of t he cource includes t he acquisit ion

• f :

❍ knowledge about t he concept s emergence, emergent

behavior and emergent syst ems;

❍ knowledge about how agent based t echniques can be used

as t ools f or modelling and simulat ion; and

❍ knowledge of what applicat ions of emergent syst ems can

be used f or and how t o evaluat e t hem.

2/11 - 04 3 Emergent Systems, Jonny Pettersson, UmU

Cource Descript ion (cont .)

❒ Moment 1, t eoridel, 3 poäng

❍ Målet med kur sen är at t ge en f örst åelse f ör

emergent a syst em. Emer gent a syst em är syst em där syst emet s bet eende uppst år som en emergent egenskap ur int erakt ionen mellan syst emet s delar. Emergent a egenskaper kan

• bser veras i alla icke-linj ära syst em som är

t illr äckligt komplexa, både nat urliga och ar t if iciella. Under kur sen kommer bland annat f r akt aler , kaos, komplexa syst em och adapt at ion at t behandlas. Kur sen ut gör en gr und f ör kursen Design av samverkande syst em. ❒ Moment 2, laborat ionsdel, 2 poäng

❍ Obligat oriska uppgif t er

2

2/11 - 04 4 Emergent Systems, Jonny Pettersson, UmU

This Cource

❒ Text book + paper s

❍ Comput er programs

❒ Focus on

❍ Fract als ❍ Chaos ❍ Complex Syst ems ❍ Adapt at ion

❒ Cont ent s

❍ Lect ures ❍ Guest lect ures ❍ Assignment s ❍ Proj ect 2/11 - 04 5 Emergent Systems, Jonny Pettersson, UmU

Assignment s and Proj ect

❒ Net Logo and t ermit es

❍ Fir st (?) cont act wit h Net Logo ❍ Termit es – a simple syst em wit h emer gent

behavior ❒ Ant Algorit hms ❒ Genet ic Algorit hms ❒ 2 and 2 in t he assignment s ❒ Proj ect

❍ 1 t o 4 in t he proj ect

2/11 - 04 6 Emergent Systems, Jonny Pettersson, UmU

Teaching

❒ Pedagogical t hought s ❒ Slides

❍ Sour ce ❍ Cont ent s ❍ Language

❒ What should you learn?

3

2/11 - 04 7 Emergent Systems, Jonny Pettersson, UmU

The rest of t oday

❒ Concept s

❍ Emergence, emer gent syst ems, …

❒ Lif e

❍ Real lif e ❍ Ar t if icial lif e

❒ Topics

2/11 - 04 8 Emergent Systems, Jonny Pettersson, UmU

Emergence

❒ Def init ions

❍ The whole is more t hen a sum of part s ❍ The global behavior could not be pr edict ed

f rom lower levels

❍ Somet hing t hat emer ges in t he int er act ions

bet ween simple(?) part s and t he environment , and t hat is not descr ibed in t he par t s ❒ Emergent propert ies ❒ Emergent syst ems ❒ Example: Ant s

2/11 - 04 9 Emergent Systems, Jonny Pettersson, UmU

Emergent Behavior

❒ Bot t om-up ❒ Dist ribut ed ❒ Local det erminat ion of behvior ❒ On all levels ❒ Example: The human body

4

2/11 - 04 10 Emergent Systems, Jonny Pettersson, UmU

Propert ies of Emer gent Syst ems

❒ Many int eract ing part s ❒ Decent ralised ❒ Non-linear ❒ Dynamic ❒ Compet it ion and cooperat ion ❒ Emergent propert ies

2/11 - 04 11 Emergent Systems, Jonny Pettersson, UmU

Many int eract ing part s

❒ Societ ies made of many people, people

made of many or gans, organs made of many cells

❒ A syst em of par t s because of int eract ions ❒ Number of part s may dif f er ❒ Massive parallelism

❍ Of t en many simple part s doing t he same t hing ❍ Complexit y comes f r om int eract ion

❒ Example: Weat her

2/11 - 04 12 Emergent Systems, Jonny Pettersson, UmU

Decent ralised

❒ Self -organisat ion

❍ The order emer ges f rom t he syst em it self

❒ Advant ages of decent ralisat ion

❍ Easier t o adapt t o changes ❍ A syst em does not need t o have a smar t leader

❒ Examples:

❍ WWW ❍ Peer-t o-peer archit ect ural models

5

2/11 - 04 13 Emergent Systems, Jonny Pettersson, UmU

Non-linear

❒ Do not obey t he superposit ion principle

❍ Out put is not propor t ional t o input

❒ I nt eract ions bet ween part s ❒ Example: Phase t ransit ions

❍ Solid – liquid - gas

2/11 - 04 14 Emergent Systems, Jonny Pettersson, UmU

Dynamic

❒ Of t en t he int er act ions cont inues on and on

❍ Does not always come t o a ”f ixed” st at e ❍ Can be bet t er t o handle changing environment s

❒ Dynamic syst ems can have dif f er ent amount s of

complexit y

❒ Example:

Societ y

2/11 - 04 15 Emergent Systems, Jonny Pettersson, UmU

Compet it ion and Cooperat ion

❒ Some agent s may cooperat e ❒ Some agent s may f ight f or t he same

resource

❒ Some agent s may dest r oy what ot hers

t ries t o do

❒ Example: Producer-consumer syst ems

6

2/11 - 04 16 Emergent Systems, Jonny Pettersson, UmU

Discussion

❒ Quest ions t o consider:

❍ What is t he emergent pr oper t y? ❍ What is t he goal of t he syst em? ❍ Does each agent know t he goal? ❍ How was t he syst em cr eat ed?

❒ Emergent syst ems in t he socit y ❒ Emergent syst ems in t he nat ure

2/11 - 04 17 Emergent Systems, Jonny Pettersson, UmU

Complex Syst ems

❒ (Almost ) like emergent syst ems

❍ Many part s ❍ I nt er dependent par t s

❒ Dif f icult t o under st and

❍ The behavior of t he whole syst em underst ood

f rom behvior of t he par t s

❍ The behavior of t he par t s depends of t he

behavior of t he whole syst em ❒ Example: Family

2/11 - 04 18 Emergent Systems, Jonny Pettersson, UmU

Adapt at ion

❒ Can lead t o improved f it ness and

perf or mance, or j ust t o be able t o survive

❒ Adapt at ion can happen in t hr ee ways

❍ I mproved handling of an event by t he agent ❍ Learning – in t he lif et ime of t he agent ❍ Evolut ion – acr oss generat ions

7

2/11 - 04 19 Emergent Systems, Jonny Pettersson, UmU

Complex Adapt ive Syst ems

❒ An complex adapt ive syst em is a syst em

consist ing of many int eract ing part s. The behavior of t he syst em emer ges out of t he parallel int eract ions bet ween t he part s and t he environment wit hout any global plan. The part s adapt and evolve over t ime.

❒ Example: Ecosyst ems

2/11 - 04 20 Emergent Systems, Jonny Pettersson, UmU

Lif e – What is Lif e?

❒ Vit alism

❍ Lif e is ”somet hing” ext ra over and above t he det ailed

• rganizat ion of a mat erial organism

❒ Langt on, 1988

❍ ”…

living organisms are not hing more t han complex biochemical machines. … A living organism … must be viewed as a large populat ion of relat ively simple machines.”

❍ ”Lif e is a propert y of f orm, not mat t er”

❒ Flake, 1998

❍ ”Nat ure appears t o be a hierarchy of comput at ional

syst ems t hat are f orever on t he edge bet ween comput abilit y and incomput abilit y.”

2/11 - 04 21 Emergent Systems, Jonny Pettersson, UmU

Lif e - Biology

❒ Does it have t o be t he st udy of coal-based

lif e?

❒ The Concise Oxf or d Dict ionary, 1990

❍ Biology – The st udy of living or ganisms

❒ Mer riam-Webst er ´ s Collegiat e Dict ionary

❍ Biology - A branch of knowledge t hat deals wit h

living or ganisms and vit al processes ❒ From greek

❍ bios – lif e ❍ logus - discour se

8

2/11 - 04 22 Emergent Systems, Jonny Pettersson, UmU

Lif e – Art if icial Lif e

❒ Langt on, 1988

❍ ”Art if icial Lif e is t he st udy of man-made

syst ems t hat exhibit behaviors char act er ist ic

• f nat ur al living syst ems.”

❍ ”…

Ar t if icial Lif e can cont r ibut e t o t heor et ical biology by locat ing lif e-as-we-know-it wit hin t he larger pict ure of lif e-as-it -could-be.”

❍ ”The ar t if icial in Ar t if icial Lif e r ef er s t o t he

component par t s, not t he emer gent pr ocesses.”

2/11 - 04 23 Emergent Systems, Jonny Pettersson, UmU

From Chaos t o Lif e

❒ Living organisms are nonlinear syst ems! ❒ Living or ganisms are complex syst ems! ❒ How has nat ure achieved t his?

2/11 - 04 24 Emergent Systems, Jonny Pettersson, UmU

From Chaos t o Lif e - Nat urally

❒ Evolut ion t hrough nat ural select ion

❍ Darwin, The Origin of Species, November 24,

1859

❍ Genot ype – phenot ype ❍ Cr it er ia f or evolut ion

• Heredit y
• Variabilit y
• Fecundit y

❒ Co-evolut ion

❍ A necessar y condit ion?

❒ Self -similarit y ❒ Self -organizat ion

9

2/11 - 04 25 Emergent Systems, Jonny Pettersson, UmU

From Chaos t o Lif e - Art if icially

❒ How t o do t his art if icially?

❍ Can not pr edict t he global behavior of simple

int eract ing subpar t s

❍ Can not decide which subpar t s t o use t o get a

predet ermined global behavior ❒ You must ”run” t he syst em t o see what kind

• f global behavior it generat e

❒ You need met hods t o search t hrough t he

solut ion space of a nonlinear syst em

2/11 - 04 26 Emergent Systems, Jonny Pettersson, UmU

From Chaos t o Lif e - Art if icially

❒ Met hods in Art if icial Lif e

❍ Lindenmayer syst ems ❍ Cellular Aut omat a ❍ Boids, her ds and f locks ❍ Ant Algor it hms ❍ Genet ic Algor it hms ❍ And mor e…

2/11 - 04 27 Emergent Systems, Jonny Pettersson, UmU

Fract als

❒ Lindenmayer syst ems

❍ Consist of set s of r ules

f or r ewr it ing st r ings of symbols

❍ “Random pr ocesses in

nat ur e ar e of t en self - similar on var ying t empor al and spat ial scales” (Flake, 1998)

❍ Example: Flake

10

2/11 - 04 28 Emergent Systems, Jonny Pettersson, UmU

Chaos – Dynamic Syst ems

❒ Examples of dynamic

syst ems

❍ Comput abilit y -

I ncomput abilit y

❍ Fract als ❍ Cellular Aut omat a 2/11 - 04 29 Emergent Systems, Jonny Pettersson, UmU

Chaos - Charact erist ics

❒ Det er minist ic

❍ Not random

❒ Sensit ive

❍ Ext remely sensit ive t o init ial condit ions

❒ Er godic

❍ A chaot ic syst em will r et ur n t o t he ”same” place

2/11 - 04 30 Emergent Systems, Jonny Pettersson, UmU

Cellular Aut omat a

❒ J ohn Von Neumann,

St anislaw Ulam, 1940s

❒ One-dimensional

❍ A linear gr id wit h cells

• f f init e-st at e-

machines

❍ At each t ime st ep, t he

next st ep of a cell is comput ed as a f unct ion

• f it s neighbors st at es

❍ Four complexit y

classes

11

2/11 - 04 31 Emergent Systems, Jonny Pettersson, UmU

Cellular Aut omat a

❒ Two-dimensional

❍ Exemple: Conway´ s Game of Lif e

• Loneliness: Less t han t wo neighbors, die
• Overcrowding: More t han t hree neighbors, die
• Reproduct ion: Empt y cell wit h t hree neighboors, live
• St asis: Exact t wo neighboors, st ay t he same

2/11 - 04 32 Emergent Systems, Jonny Pettersson, UmU

Aut onomous Agent s

❒ Ant Algorit hms

❍ Ant s deposit s

pheromones when t hey move

❍ They dynamically f ind

t he ”shor t est ” way

2/11 - 04 33 Emergent Systems, Jonny Pettersson, UmU

Aut onomous Agent s

❒ Boids, her ds and f locks

❍ Craig Reynolds, 1986 ❍ Simple behaviors Separation Cohesion Alignment Neighborhood

12

2/11 - 04 34 Emergent Systems, Jonny Pettersson, UmU

Genet ics and Evolut ion

❒

Genet ic Algorit hms

❍

Holland, 1960s

❍

A simple algor it hm:

• 1. St art wit h a randomly generat ed populat ion of

candidat e solut ions t o a problem

• 2. Calculat e t he f it ness of each solut ion in t he

populat ion

• 3. Apply select ion and genet ic operat ors t o t he

populat ion t o creat e a new populat ion

• 4. Go t o st ep 2

❍

Example: Br eve, J on Klein

2/11 - 04 35 Emergent Systems, Jonny Pettersson, UmU

Compet it ion and Cooperat ion

❒ Game t heory ❒ The evolut ion of cooperat ion ❒ The Prisoner ’s Dilemma

❍ I t erat ed ❍ Spat ial

2/11 - 04 36 Emergent Systems, Jonny Pettersson, UmU

Simulat ion and Modelling

❒ Emergence is explained ❒ Want t he simplest syst em t hat produce

t he emergent behavior

❍ Ockham’s razor (1285 – 1347/ 49)

❒ Not complet e models of realit y

❍ Focus on t he essence of t he syst em ❍ Not clones

13

2/11 - 04 37 Emergent Systems, Jonny Pettersson, UmU

Why St udy Emer gent Syst ems?

❒ Fundament al t o t heory and implement at ion

• f massively parallel, dist ribut ed

comput at ion syst ems

❒ Goals/ challanges

❍ Ef f iciency ❍ Self -opt imising ❍ Adapt ive ❍ Robust t o f ailures ❍ Secur it y

❒ Nat ure as a source

❍ Tr y t o under st and nat ur e ❍ Use nat ur e as an inspir at ion

2/11 - 04 38 Emergent Systems, Jonny Pettersson, UmU

Summar y

❒ Concept s

❍ Emergence, emer gent syst ems, …

❒ Lif e

❍ Real lif e ❍ Ar t if icial lif e

❒ Topics

2/11 - 04 39 Emergent Systems, Jonny Pettersson, UmU

Next Time

❒ Fract als ❒ Net Logo ❒ Assignment 1