adaptive traffic and dynamical networks: from plants to people methods: agents (i.e. decision-making...

18
Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical + simulation + analytics Empirical: Dynamical evolution of nutrient networks in fungal and slime-mold systems AVMs and supply networks in the brain Angiogenesis in cancer tumour growth Virus spreading on networks Supply chains, flow of cases in judicial systems Flow of rumours around FX currency markets Issues: Flow of objects on network: group formation & crowding, congestion Feedback onto network structure: structure vs function What is ‘best’ ? optimal, fault tolerant or just ‘good enough’ centralized vs decentralized competitive vs. cooperative How to control, manage, design ? http://sbs-xnet.sbs.ox.ac.uk/complexity/ complexity_splash_2003.asp Mark Fricker, Paul Summers, Pak Ming Hui, Charley Choe, David Smith, Chiu Fan Lee, Tim Jarrett, Sean Gourley, Neil Johnson

Upload: bryan-crawford

Post on 06-Jan-2018

227 views

Category:

Documents


1 download

DESCRIPTION

Constructing an agent-based fungus.. so function drives structure

TRANSCRIPT

Page 1: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

Adaptive Traffic and Dynamical Networks: From Plants to People

• Methods:agents (i.e. decision-making particles) + networks (i.e. connectivity)empirical + simulation + analytics

• Empirical:Dynamical evolution of nutrient networks in fungal and slime-mold systemsAVMs and supply networks in the brainAngiogenesis in cancer tumour growthVirus spreading on networksSupply chains, flow of cases in judicial systemsFlow of rumours around FX currency markets

• Issues:Flow of objects on network: group formation & crowding, congestionFeedback onto network structure: structure vs functionWhat is ‘best’ ?

optimal, fault tolerant or just ‘good enough’centralized vs decentralizedcompetitive vs. cooperative

How to control, manage, design ?

http://sbs-xnet.sbs.ox.ac.uk/complexity/ complexity_splash_2003.asp

Mark Fricker, Paul Summers, Pak Ming Hui, Charley Choe, David Smith, Chiu Fan Lee, Tim Jarrett, Sean Gourley, Neil Johnson

Page 2: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

NC

P

C

N

Page 3: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

Constructing an agent-based fungus

. . sofunctiondrives structure

Page 4: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

Organism ‘develops’ ability to walk/hunt/forageIssues: localization-delocalization, efficient vs. adaptable

QuickTime™ and a decompressor

are needed to see this picture.

Page 5: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

. . .should I try a short-cutthrough the centre?

Page 6: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

If cost c of using central hub is non-linear, we find:Abrupt changes in optimal network structure, which are induced by small changes in c

Diverse set of structurally inequivalent networks, which are functionally equivalent

Page 7: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

centralized vs decentralized ? Phys. Rev. Lett. 2005

Page 8: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

If cost c of using central hub is non-linear, we find:Abrupt changes in optimal network structure, which are induced by small changes in c

Diverse set of structurally inequivalent networks, which are functionally equivalent

Page 9: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

Efficient, but deadly . . . .

cancer: angiogenesis brain: AVM, aneurysm

Page 10: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

updated history at time t +1

. . . . 1 0

n−1 t[ ]

S

history at time t

. . . . 0 1agent memory m = 2

action+1

action-1

n+1 t[ ]

f d

b e

c

a global outcome0 or 1

f d

b e

c

a

strategyspace

Page 11: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

Evolutionary Minority Game (EMG) Phys. Rev. Lett. ‘99

agent’s strategy/‘gene’ p mutates

time: tτ

agent’s performance or ‘wealth’

time

kp0 10.5kp0 10.5

dist

ribut

ion

of a

gent

s

0t=

0.5

steady stateself-organized segregation into

Anti-crowds and Crowds

p = probability agent follows common information: past history, news, rumor (right or wrong)

p p

Page 12: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

coin-toss

basic Minority Game (MG)crowding/congestion reduced by- acting ‘dumb’- choosing second-best- mis-information- heterogeneity in abilities, e.g. m

agent memory m

hybrid

EMG

typical fluctuation size

Evolutionary Minority Game (EMG) Phys. Rev. Lett. ‘99

Page 13: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

General global resource level L (e.g. # seats)A(t)

L N = 100 agents wish to access resource (e.g. attend bar)

L

‘freezing’of evolution

A + ΔA < L

time: tτ

A(t) = L

A(t)

A

ΔA

Page 14: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

system’s time evolution

Phys. Rev. E (2004)

Distribution of duration ofextreme large changes in

variable-N evolutionary MG

-4

∆ t

largest changes/crashes are ‘different’ -7

QuickTime™ and aBMP decompressor

are needed to see this picture.

Page 15: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

history at time t+1

. . . . 1 0

n−1 t[ ]

S

agent memory m = 2

S

n+1 t[ ]

f d

b e

c

a

2m

22m

histories

strategies

11 10 01 00

-1 -1 -1 -1

-1 -1 -1 +1 . . . .

-1 -1 +1 +1 . . . .

-1 +1 +1 +1 . . . .

+1 +1 +1 +1

f d

b e

c

a

action+1

action-1

history at time t

. . . . 0 1

global outcome0 or 1

Page 16: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

memory m

volatility

Memory m 2m+1 << N.s 2m+1 ~ N.s 2m+1 >> N.sCrowd

sizelarge medium ~ 1

Anticrowdsize

small medium ~ 0

Net crowd –anticrowdpair size

large>> 1

small small~ 1

# crowd -anticrowd

pairs

~ 2m

<< N~ 2m

< N< 2m

~ N

random

crowd - anticrowd pairs executeuncorrelatedrandom walks

sum of variances

walk step-size

# of walks

typical fluctuation sizeMinority Game:

Each agent has s=2,3,4,.. strategieswith memory-length m

2mhistories

11 10 01 00

-1 -1 -1 -1

-1 -1 -1 +1 . . . .

-1 -1 +1 +1 . . . .

-1 +1 +1 +1 . . . .

+1 +1 +1 +1

Page 17: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

Hub Capacity L=40

Add in agent decision-making . . . .

Page 18: Adaptive Traffic and Dynamical Networks: From Plants to People Methods: agents (i.e. decision-making particles) + networks (i.e. connectivity) empirical

K4