SelfOrganisation in Specknets

1
Self-Organisation in Specknets

• Tom ten Thij
• Napier University
• specknet_at_tomtenthij.nl

2
Self-organisation

• Only localized interactions between elements
• Positive feedback loops can not be avoided
• Too large-scale for any form of global control

3
Amorphous Computing Project (MIT)

• Goal engineering of global structure in terms of
  local unsynchronized interactions
• Or engineering of emergent order
• Aimed at developing programming methodologies for
  systems composed of vast numbers of identical
  agents

4
A methodology gradient wave

• Source emit message with counter 0
• Recipients remember counter value and emit
  message with increased counter
• Messages with higher counter values are ignored
• Smoothing

Image source Amorphous computing project
5
Inspiration from fruit flies

• Morphogen gradients / Position
• Local inhibition

Images source The Interactive Fly
6
How to form a tube

• Assuming
• Gradient wave is only source of locational
  information
• Only local undirected interaction
• Two poles are established
• ?

Images source Amorphous computing project
7
Tube formation

• Phase 2
• - Gradient

• Phase 1
• Gradient

• Phase 3
• Form tube

• Phase 4
• Smooth

Images source Amorphous computing project
8
Tube formation movie
Movie source Amorphous computing project
9
Chosen problem topology feature detector

• Examples
• Phallic-ness
• Circularity
• Path fork
• Why this problem?
• Relatively simple but still hard to do
  distributively
• Many possible solutions can be envisioned
• Optimal solutions hard to design due to abundance
  of interactions (chaos)

c
O
Y
10
Sketch of a solution for circularity detection

• Use local inhibition to define at least 2 sources
  (a certain hop-count-distance apart)
• Diffuse gradients from sources, other specks
  propagating only first gradient that reaches them
• A border is where to gradients meet. For every
  border elect one b!speck using lateral inhibition
• Each b!speck broadcasts its id over the network
  so all specks can count the number of borders
• If b!speck ? source ? circle
• else ? phallic

11
Optimization by evolutionary algorithms

• First hand design several property detectors
• Search space
• Parameters of packet propagation and state
  transisions
• Fitness
• Time before correctly detecting property
• Robustness of detection
• Hopefully optimisation that is hard to
  hand-design can be found

12
The End

• Feedback / Questions?

13
References

• Amorphous computing
• http//www.swiss.ai.mit.edu/projects/amorphous/
• Radhika Nagpal. A Catalog of Biologically-Inspired
  Primitives for Engineering Self-Organization. In
  Engineering Self-Organising Systems, 2003.
• The Interactive Fly
• http//www.sdbonline.org/fly/aimain/1aahome.htm