View
217
Download
1
Tags:
Embed Size (px)
Citation preview
Emergence: the relation between static network structure and the dynamic qualities of landscapes
Jeremy Yamashiro, University of UtahJonathan Butner, University of UtahChase Dickerson, University of UtahThomas Malloy, University of Utah
How does static network structure
relate to the dynamic flow of
information across that structure?
TILE IMAGE
Research Questions
•How do different Generating Rules structure different networks?
•How does network structure affect the qualities of its emergent landscapes?
Components of an NK Boolean Network•static network: binary nodes and
wiring
•attractor landscapes
A B
C D
The flow of state vectors over time falls into
attractors, which we may think of as basins in a
landscape
IN OUTA 2 2B 2 1C 2 2D 2 3
Generation rules
•Simultaneous Random (SR)
All (N) nodes are inserted and form links simultaneously; each node has equal probability of taking one of its K inputs from any other node
•Ordered Random (OR)
(N) nodes are inserted one at a time, and each forms all K inputs before the next node is inserted
• Simultaneous• Each Randomly connects to
2 other nodes
• Ordered• Each Randomly connects to 2
other nodes• But only connect to those before
it
Generation Rule determines how nodes are connected to each
other
Types of NetworkVery different network structures emerge
from different Generation Rules
Ordered RandomSimultaneous Random
Network BehaviorBoolean landscapes: attractors/basins
Nodes 1-100{
State Vector at Time T to T+NL=12
Ruggedness
• The relative number of attractors a network can produce
• Ruggedness is an indicator of the behavioral complexity of a system; greater ruggedness => greater behavioral diversity, flexibility, adaptibility
Attractor Homogeneity
• Degree to which attractor lengths (L’s) are limited or proliferate
• Attractor homogeneity indicates consistency between attractors; a homogenous system can be more coherent, despite high ruggedness, than a heterogenous system
Analysis of Network Structure
•Generated 1000 SR and 1000 OR networks, N=100, K=3.
•log-log slope of regression line of nodes/output links distribution of each network
Analysis of Network Structure
.95 confidence interval for mean log-log regression slopes of nodes/output links distribution.
SR OR
-1.0314+/-0.0194
-1.4865+/- 0.2485
Landscapes as a function of
Generation Rule
•ruggedness
•attractor homogeneity
We generated 50 SR and 50 OR at N=100, K=3 for landscape analysis.
Ruggedness as Function of
Generation RuleMean number of basins
SR OR
859.34 Standard Deviation = 249.11
970.48 Standard Deviation = 94.27
Attractor Homogeneity as Function of
Generation RuleMean number of attractor cycle lengths (L’s)
SR SO
24.54Standard Deviation = 21.16
2.02 Standard Deviation
= 0.77
Landscapes as Function of Log-
Log Slope•We pooled the 2 samples of 50 networks of each Generating Rule and pooled them into 1 sample of 100 networks
•regression analyses of log-log slope and ruggedness, and log-log slope and attractor homogeneity
Landscapes as Function of Log-
Log SlopeLog-log slope fails to predict ruggedness (total number of attractors)
Summary•Ordered Random (OR) Generation Rule
produces networks with steeper (more negative) log-log slope, within the fractal range (Butner, Pasupathi, Vallejos, 2008).
•OR networks produce more rugged landscapes than SR networks.
•OR/fractal networks produce more homogenous attractor landscapes
DiscussionMapping:
Boolean Network => Neural Net
Attractor Thought=>
Stream of Cognition/Thought
Attractor Landscapes=>
Discussion• Ruggedness and attractor homogeneity are a
set of freedoms and constraints on a system’s behavior
• Greater ruggedness means the diversity of thoughts potential to a system is very high, even while homogenous attractor landscapes may produce greater coherence between different thoughts or sets of thoughts.
• The landscapes at the intersection of high ruggedness and high attractor homogeneity allow for behavior that is simultaneously highly diverse (creative?) and internally consistent
Discussion
Fractal-like networks produce attractor landscapes of great consistency and richness; fractal-like wiring of the neural net may support more complex and coherent cognitive processes.
Acknowledgments
This project was supported in part by a grant from the University of Utah’s Undergraduate Research Opportunities program.
ReferencesBarabási A. L. & Réka A. (1999). Emergence of Scaling in Random Networks. Science 286, 509 - 512
Barabási, A.L. (2003) Linked. Cambridge: Plume. Bateson, G. (2002) Mind and Nature. Cresskill: Hampton Press, Inc. (originally published by Dutton, 1979).
Butner, J., Pasupathi, M., Vallejos, V. (2008). When the facts just don’t add up: The fractal nature of conversational stories. Social Cognition, 26, 670-699.
Erdos, P. & Rényi, A. (1959) On Random Graphs I. Publ. Math. Vol ?, 290–297. Erdos, P. & Rényi, A. (1960). On the evolution of random graphs. Publications of the Matkemafical Insfifufe of the Hungarian Academy of Sciences, 5.
Kauffman, S. A. (1969). Metabolic stability and epigenesis in randomly connected nets. Journal of Theoretical Biology, 22, 437.
Kauffman, S. A. (1971). Gene regulation networks. Current Topics in Developmental Biology,6, 145.
Kauffman, S. A. (1993). The origins of order: self-organization and selection in evolution.Oxford: Oxford University Press.
Malloy, T. E., Bostic St Clair, C. & Grinder, J. (2005). Steps to an ecology of emergence. Cybernetics & Human Knowing, 12, 102-119.
Malloy, T.E., Butner, J., & Jensen, G. C. (2008). The emergence of dynamic form through phase relations in dynamic systems. Nonlinear Dynamics, Psychology, and Life Sciences, 12, 371-395.
Malloy, T.E., Jensen, G.C. (2008). Dynamic constancy as a basis for perceptual hierarchies. Nonlinear Dynamics, Psychology, and Life Sciences, 12, 191-203.
Malloy, T. E., Jensen, G. C., & Song, T. (2005) Mapping knowledge to Boolean dynamic systems in Bateson’s epistemology. Nonlinear Dynamics, Psychology, and Life Sciences, 9, 37- 60.
Mitchell, M. (2009). Complexity. Oxford: Oxford University Press.
Turing,A.M.(1952).The chemical basis of morphogenesis. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences, 237, 37-72.
Watts, D. J., & Strogatz, S. H. (1998). Collective dynamics of 'small world' networks. Nature 393, 440-42.