8
Fitness Landscape Fitness Landscape Simulation Simulation Modeling Architecture and Modeling Architecture and Component Research Component Research Laurent Mirabeau Igor Feldman Murat Cokol Ezra Goodnoe

Fitness Landscape Simulation

Embed Size (px)

DESCRIPTION

Fitness Landscape Simulation. Modeling Architecture and Component Research. Ezra Goodnoe. Murat Cokol. Igor Feldman. Laurent Mirabeau. Innovation Theory. Dominant Design After a given time a few dominant designs will emerge. They can be thought as basin of attractions. Dominant Design. - PowerPoint PPT Presentation

Citation preview

Page 1: Fitness Landscape Simulation

Fitness Landscape Fitness Landscape SimulationSimulation

Modeling Architecture and Modeling Architecture and Component ResearchComponent Research

Laurent Mirabeau

Igor FeldmanMurat Cokol

Ezra Goodnoe

Page 2: Fitness Landscape Simulation

Innovation TheoryInnovation Theory

• Dominant Design– After a given time a few dominant designs will

emerge. They can be thought as basin of attractions

Dominant Design

Product

Process

Page 3: Fitness Landscape Simulation

Innovation TheoryInnovation Theory

• Architecture Vs Component Innovation– Companies organize around the architecture of

the product they research / manufacture

Page 4: Fitness Landscape Simulation

The ModelThe Model

• One central hub: architecture team• Several peripheral nodes: component teams• Hub decisions impact overall system• Generic network motif

Page 5: Fitness Landscape Simulation

Fitness FunctionFitness Function

ji

ssjisj

si

i

si

si LccccF

,',,,

')(

Create n nodes, initialize values 1-5 utility

Create n*(n-1)*s*s links, initialize random -1<<1

Calculate fitness for points in landscape

Random drop on landscape

Select one-mutant neighbor

Calculate fitness and walk if higher

Iterate, repeat for many agents over several landscapes

Page 6: Fitness Landscape Simulation

ResultsResults

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 20 40 60 80 100

Fitness

Iterations

10 Landscapes

100 Agents

500 Iterations

Page 7: Fitness Landscape Simulation

Emergence of Dominant DesignsEmergence of Dominant Designs

Architecture Vs Component

0

0.05

0.1

0.15

0.2

0.25

0.3

1 2-5 6-10 11-20 21-50 51-100 101-500

Iterations

Pro

babi

lity

Arc

Comp

Page 8: Fitness Landscape Simulation

Next StepsNext Steps

• Increase complexity of model by adding dependencies

• Try various strategies – Stochastic process to melt off peaks, Concurrent agents, Increase hamming distance

• Combine several motifs to create new class of problems