1 Self -Or ganizat ion - St igmergy - Advant ages Charact erist - - PDF document

1
SMART_READER_LITE
LIVE PREVIEW

1 Self -Or ganizat ion - St igmergy - Advant ages Charact erist - - PDF document

Last t ime Self -Organizat ion Cellular aut omat a Pat t ern One-dimensional A par t icular, or ganized ar rangement of obj ect s Wolf r ams classif icat ion in space or t ime Langt ons lambda par amet er I nt


slide-1
SLIDE 1

1

19/11 - 04 1 Emergent Systems, Jonny Pettersson, UmU

Last t ime

❒ Cellular aut omat a

❍ One-dimensional ❍ Wolf r am’s classif icat ion ❍ Langt on’s lambda par amet er ❍ Two-dimensional

  • Conway’s Game of Lif e

❒ Pat t ern f ormat ion in slime molds

❍ Dict yost elium discoideum ❍ Modeling of pat t er n

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

Out line f or t oday

❒ Self -Organizat ion ❒ Aut onomous Agent s ❒ Real Ant s ❒ Virt ual Ter mit es ❒ Virt ual Ant s ❒ Ant Algorit hms

19/11 - 04 3 Emergent Systems, Jonny Pettersson, UmU

Self -Organizat ion

❒ ”Self -organizat ion is a process in which

pat t ern at t he global level of a syst em emer ges solely f rom numer ous int eract ions among t he lower-level component s of t he syst em. Moreover, t he r ules specif ying int eract ions among t he syst em’s component s are execut ed using only local inf ormat ion, wit hout ref erence t o t he global pat t ern.” – Camazine et al, p. 8

19/11 - 04 4 Emergent Systems, Jonny Pettersson, UmU

Self -Organizat ion

❒ Pat t ern

❍ A par t icular, or ganized ar rangement of obj ect s

in space or t ime ❒ I nt eract ions

❍ Based on local inf or mat ion only - no global

inf ormat ion

❍ P

hysical laws

❍ Genet ically cont r olled proper t ies of t he

component s

19/11 - 04 5 Emergent Systems, Jonny Pettersson, UmU

Self -Or ganizat ion - I ngredient s

❒ Posit ive f eedback

❍ Act ivit y amplif icat ion

❒ Negat ive f eedback

❍ Act ivit y balancing

❒ Amplif icat ion of random f luct uat ions ❒ Mult iple int eract ions

19/11 - 04 6 Emergent Systems, Jonny Pettersson, UmU

Self -Or ganizat ion - I nf ormat ion

❒ Signals

❍ St imuli shaped by nat ural select ion specif ically

t o convey inf ormat ion ❒ Cues

❍ St imuli t hat convey inf ormat ion only

incident ally ❒ Gat hered f rom one’s neighbors

❍ St imuli-r esponse, simple behavioral r ules of

t humb ❒ Gat hered f r om work in progr ess

❍ St igmer gy ❍ Random f luct uat ion and chance het er ogeneit ies

slide-2
SLIDE 2

2

19/11 - 04 7 Emergent Systems, Jonny Pettersson, UmU

Self -Or ganizat ion - Charact erist ics

❒ Dynamic syst ems ❒ Exhit emer gent propert ies

❍ At t ract or s ❍ Mult ist abilit y ❍ Bif ur cat ions ❍ P

ar amet er t uning

❍ Environment al f act ors

❒ Adapt ive syst ems ❒ Dif f erent pat t erns may result f rom t he

same mechanism

❒ Simple rules, complex pat t erns

19/11 - 04 8 Emergent Systems, Jonny Pettersson, UmU

Self -Or ganizat ion – Alt ernat ives

❒ Cent ral leader

❍ Need ef f ect ive communicat ion and cognit ive

abilit ies ❒ Blueprint s

❍ Most be st ored

❒ Recipes

❍ Hinder s f lexibilit y

❒ Templat es

❍ Must be avaiable

19/11 - 04 9 Emergent Systems, Jonny Pettersson, UmU

St igmergy

❒ A recursive cont rol syst em ❒ Ef f ect ive f or coor dinat ion in space and

t ime

❒ A sequence of qualit at ively dif f erent

st imulus-response behavior s

❒ Two t ypes:

❍ Qualit at ive st igmergy ❍ Quant it at ive st igmer gy

19/11 - 04 10 Emergent Systems, Jonny Pettersson, UmU

St igmergy - Advant ages

❒ Permit e simpler agent s ❒ Decrease direct communicat ion bet ween

agent s

❒ I ncrement al improvment ❒ Flexible, since when environment changes,

agent s respond appropriat ely

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

Aut onomous Agent

❒ ”a unit t hat int eract s wit h it s environment

(which probably consist s of ot her agent s)

❒ but act s independent ly f rom all ot her

agent s in t hat it does not t ake commands f rom some seen or unseen leader ,

❒ nor does an agent have some idea of a

global plan t hat it should be f ollowing.”

  • Flake, p. 261

19/11 - 04 12 Emergent Systems, Jonny Pettersson, UmU

Real Ant s

❒ I magine if art if icial syst ems could do t he

t hings ant s can do?

❒ Why ant s?

❍ Amazonas: 30% of biomass is ant s/ t ermit es ❍ Amazonas: dr y weight of social insect s is f our

t imes t hat of ot her land animals

❍ Eart h: ~10% of t ot al biomass (like humans)

slide-3
SLIDE 3

3

19/11 - 04 13 Emergent Systems, Jonny Pettersson, UmU

Army Ant s

❒ 100 000s in colony ❒ Cr eat e t empor ar y

”bivouacs”

❒ Act like unif ied ent it y

(Pictures from AntColony.org) 19/11 - 04 14 Emergent Systems, Jonny Pettersson, UmU

Fungus-Gr owing Ant s

❒ "A Leaf Cut t er Colony

can st r ip t he t allest

  • f t r ees in a single
  • day. Equivalent

consumpt ion of a f ull grown cow in t he same t ime!"

❒ ”Cult ivat e” f ungi

under ground

❒ Fert ilize wit h

compost f rom chewed leaves

(Pictures from AntColony.org) 19/11 - 04 15 Emergent Systems, Jonny Pettersson, UmU

Fungus Cult ivat or Nest

(Picture from AntColony.org) 19/11 - 04 16 Emergent Systems, Jonny Pettersson, UmU

Harvest er Ant s

❒ Find shor t est pat h t o

f ood

❒ Pr iorit ize f ood sour ces

based on dist ance and ease of access

(Picture from The Texas A&M University System) 19/11 - 04 17 Emergent Systems, Jonny Pettersson, UmU

Adapt ive P at h Opt imizat ion

19/11 - 04 18 Emergent Systems, Jonny Pettersson, UmU

Virt ual Termit es

❒ The assigment ❒ Why does t he number of piles decrease? ❒ How t o improve t he perf or mance wit h t wo

t ype of t ermit es and t wo t ype of chips?

❒ How does dest royers af f ect t he syst em?

slide-4
SLIDE 4

4

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

Langt on’s Virt ual Ant s

❒ Grid wit h whit e or black squares ❒ Virt ual ant s can f ace N, S, E, W ❒ Behavioral rule:

❍ Take a st ep f orwar d ❍ if on a whit e square then paint it black and t urn

90º right

❍ if on a black squar e then paint it whit e and t urn

90º lef t

19/11 - 04 20 Emergent Systems, Jonny Pettersson, UmU

Virt ual Ant s - Example

19/11 - 04 21 Emergent Systems, Jonny Pettersson, UmU

Virt ual Ant s – Time Reversibilit y

❒ Virt ual ant s are t ime-rever sible ❒ But , t ime-reversibilit y does not imply

global simplicit y

❒ Even a single virt ual ant int eract s wit h it s

  • wn prior hist ory

❒ Demonst rat ion

19/11 - 04 22 Emergent Systems, Jonny Pettersson, UmU

Virt ual Ant s - Conclusion

❒ Even simple, reversible local behavior can

lead t o complex global behavior

❒ Such complex behavior may creat e

st r uct ures as well as apparent ly random behavior

19/11 - 04 23 Emergent Systems, Jonny Pettersson, UmU

Ant Algorit hms

❒ Ant colony opt imizat ion (ACO) ❒ Developed in 1991 by Dorigo (PhD

dissert at ion) in collaborat ion wit h Colorni and Maniezzo

19/11 - 04 24 Emergent Systems, Jonny Pettersson, UmU

Summar y

❒ Self -Organizat ion ❒ Aut onomous Agent s ❒ Real Ant s ❒ Virt ual Ter mit es ❒ Virt ual Ant s ❒ Ant Algorit hms

slide-5
SLIDE 5

5

19/11 - 04 25 Emergent Systems, Jonny Pettersson, UmU

Next t ime

❒ Flocks, Herds, and Schools ❒ Boids