Complex Adaptive Systems

1

• Complex Adaptive Systems
• Stephen Fulwider
• COT 4810
• 19 April 2008

2
Outline

• Introduction
• Bumblebees
• Example
• Setup
• Methods
• Results
• Demo

www.anatomyacts.co.uk
3
Complex Adaptive Systems

• System of Agents
• Emergent Behavior
• Adaptation to Environment
• Can be Interconnected
• Derived from Biology

www.wikipedia.com
4
Bumblebee Thermoregulation

• Regulate hive temperature
• Varied tasks
• Flap wings
• Cluster
• Variation key to stability
• Activation temperature
• Variable time
• Specialists may emerge
• Not required, though

5
Example

• Pointer and Target System
• Agents 'push'
• Pointer towards Target
• Agents can
• Adapt some settings
• Evolve some settings
• Rates of adaptation/evolution
• can vary

6
Agents

• Direction N,E,S,W
• Weight 0,1
• Speed 0,1
• CIP 0,1

7
Adaptation

• Every Agent
• Adapt weight
• Adjust 'correctly'
• with prob. CIP
• May be incorrect
• That's okay!

8
Evolution

• Steady State GA
• 1 Reproduction / Generation
• Tournament Selection
• Fitness
• Euclidean Distance
• Winners Compared
• Better (possibly) cloned
• Better (definitely) mutated

www.cosmicvolunteers.net
9
Mutation

• Direction (random)?
• Speed (Gaussian)?
• CIP (Gaussian)?
• Weight ( Static Value)?
• Independent Mutation

www.virtualsciencefair.org
10
Population Loss

• Loss of half the population
• Occurs halfway through run
• Simulates large loss of information
• Testing how well system can recover
• We assume it will do worse
• But by how much?

11
Paths

• Random
• Vertical Line
• Circle
• Square
• Delayed Cross
• 400x400 Grid
• Max Dist 565

12
GA Wins
13
Pop Loss CAS Handles Well
14
CIP 0.50 Bad, 0.75/1.00 Similar
15
Vertical Line Path (NO POP LOSS)?
16
Vertical Line Path (YES POP LOSS)?
17
Vertical Line Path (GA ON)?
18
Square Path (GA ON)?
19
Homework Questions

• What is one defining characteristic of a complex
  adaptive system?
• What is one reason that a GA working in tandem
  with a CAS could do well?

20
References

• 1. Honey Bee Nest Thermoregulation Diversity
  Promotes Stability. Jones, Julia C., et al.
  s.l.  Science, 2004, Vol. 305.
• 2. Collective control of nest parameters in
  bumblebee colonies. Weidenmuller, Anja,
  Kleineidam, Christoph and Tautz, Jurgen. s.l. 
  Animal Behaviour, 2002, Vol. 63.
• 3. Designing Emergent Behaviors From Local
  Interactions to Collective Intelligence. Mataric,
  Maja J. s.l.  From Animals to Animats, 1992,
  Vol. 2.
• 4. Uniform Crossover in Genetic Algorithms.
  Syswerda, Gilbert. s.l.  International
  Conference on Genetic Algorithms, 1989.
• 5. Genetic Algorithms for Tracking Changing
  Environments. Cobb, Helen G. and Grefenstette,
  John J. s.l.  International Genetic Algorithms
  Conference, 1993.
• 6. Effects of population size upon emergent group
  behavior. Baray, Cristobal. IV, s.l.  Complexity
  International, 1998, Vol. 6.
• 7. MASON A New Multi- Agent Simulation Toolkit.
  Luke, Sean, et al. s.l.  Proceedings of the 2004
  SwarmFest Workshop, 2004.
• 8. Simulation meta-architecture for analyzing the
  emergent behavior of agents. Schaefer, Lisa A.,
  et al. s.l.  Complexity International, Vol. 10.
• 9. Collaborative Control of Robot Motion
  Robustness to Error. Ken Goldberg, Billy Chen.
  Maui, Hawaii  International Conference on
  Intelligent Robots and Systems, 2001.
  0-7803-6112-3.