Bio-inspired Networking and Complex Networks: A Survey
Sheng-Yuan Tu
1
Outline
Challenges in future wireless networks
Bio-inspired networking
Example 1: ant colony
Example 2: immune system
Complex networks
Network measures
Network models
Phenomena in complex networks
Dynamical processes on complex networks
Further research topics
2
Challenges in Future Wireless Networks
Scalability
By 2020, there will be trillion wireless devices [1] (e.g. cell phone, laptop, health/safety care se nsors, …)
Adaptation
Dynamic network condition and diverse user demand
Resilience
Robust to failure/malfunction of nodes and to intru ders
3
Bio-inspired Networking
Biomimicry: studies designs and processes in n ature and then mimics them in order to solve h uman problems [3]
A number of principles and mechanisms in large scale biological systems [2]
Self-organization: Patterns emerge, regulated by fe edback loops, without existence of leader
Autonomous actions based on local information/inter action: Distributed computing with simple rule of t humb
Birth and death as expected events: Systems equip w ith self-regulation
Natural selection and evolution
Optimal solution in some sense
A special issue on bio-inspired networking wil l be published in IEEE JSAC in 2
ndquarter 2010 .
4
Bio-inspired Networking
Observatio n, verbal description Observatio n, verbal description
Math. Model (Diff. eq.,
prob.
methods, fuzzy logic,
…)
Math. Model (Diff. eq.,
prob.
methods, fuzzy logic,
…) Verification
,
hypothesis testing Verification
,
hypothesis testing Parameter
evaluation, prediction Parameter evaluation,
prediction
Entities mapping Entities mapping
Algorithm establishm
ent
Algorithm establishm
ent
Performan ce evaluation Performan
ce evaluation
Biological Modeling Engineering Applying
Parameter tuning Parameter
tuning
5
Example 1: Foraging of Ant Colony
Stigmergy: interaction between ants is built o n trail pheromone [6]
Behaviors [6]:
Lay pheromone in both directions between food sourc e and nest
Amount of pheromone when go back to nest is accordi ng to richness of food source (explore richest reso urce)
Pheromone intensity decreases over time due to evap oration
Stochastic model (no trail-laying in backward) :
i
i i i
dC q P fC dt
1
( )
( )
n i
i m
n i j
P k C
k C
C1
C2
Cm
P1
P2
Pm
6
Example 1: Foraging of Ant Colony
Parameter evaluat ion:
Ω: flux of ants
q: amount of pher omone laying
f: rate of pherom one evaporation
k: attractiveness of an unmarked pa th
n: degree of nonl inearity of the c hoice
Shortest path sea rch
1 2
/ 2 ( m)
R q fk q q q m
[5]
7
Example 1: Foraging of Ant Colony
Application in ad-hoc network routing [4]
Modified behaviors
Probabilistic solution construction without forward pheromone updating
Deterministic backward path with loop elimination a nd pheromone updating
Pheromone updates based on solution quality
Pheromone evaporation (balance between exploration and exploitation)
8
Example 1: Foraging of Ant Colony
Algorithm
Initiation
Path selection
Pheromone update
More other applications can be found in swarm intelligence [7].
0 /
ij m Cnn
[ ] [ ] [ ] [ ]
ik
ij ij
k ij
il il
l N
p
ij 1/ dij(1 )
ij ij
1
m k
ij ij ij
k
1/ if arc( , ) belongs to 0 otherwisek k
k ij
C i j T
9
Example 2: Immune System
Functional architecture of the IS [8]
Physical barriers: skin, mucous membranes of digest ive, respiratory, and reproductive tracts
Innate immune system: macrophages cells, complement proteins, and natural killer cells against common p athogen
Adaptive immune system: B cells and T cells
B cells and T cell are created from stem cells in the bo ne marrow ( 骨髓 ) and the thymus ( 胸腺 ) respectively by rearrangement of genes in immature B/T cells.
Negative selection: if the antibodies of a B cell match any self antigen in the bone marrow, the cell dies.
Self tolerance: almost all self antigens are presented i n the thymus.
Clonal selection: a B cell divides into a number of clon es with similar but not strictly identical antibodies.
Danger signal: generated when a cell dies before be gin old
10
Example 2: Immune System
Procedure
Antibodies of B cell match antigens (signal 1b) Antibodies of B cell matchantigens (signal 1b)
Antibodies of T cell binds the antigens (signal 1t) Antibodies of T cell binds
the antigens (signal 1t) Matching
>
Threshol d?
Clonal selection Clonal selection
Receive signal
2t?
Match antigens
? Antigen Presenting
Cell Antigen Presenting
Cell No
Yes
Yes
Yes
Danger Signal
Signal 2t
T cell sent signal 2b to B cell
T cell sent signal 2b to B cell
11
Example 2: Immune System
Application in misbehavior detection in mobile ad-hoc networks with dynamic source routing (D SR) protocol [8]
Entity mapping:
Body: the entire mobile ad-hoc network
Self-cells: well behaving nodes
Non-self cells: misbehaving nodes
Antigen: sequence of observed DSR protocol events i n the packet headers
Antibody: A pattern with the same format of antigen
Chemical binding: matching function
Bone marrow: a network with only certified nodes
Negative selection: antibodies are created during a n offline learning phase
12
Complex Networks
The above approach is more or less heuristic a nd is based on trial and error. What is theor etical framework to understanding network beha viors?
Network measures
Degree/connectivity (k)
Degree distribution
Scale-free networks
Shortest path
Six degrees of separation (S. Milgram 1960s)
Small-world effect
Clustering coefficient (C)
Average clustering coefficient of all nodes with k links C(k)
[12]
( ) (2< 3) P k k
3 # of triangles
# of connected triples of vertices
C
13
Complex Networks
Network models
Random graphs (ER model)
Start with N nodes and connect each pair of nodes with p rob. p
Node degrees follow a Poisson distribution
Generalized random graphs (with arbitrary degree di stribution)
Assign ki stubs to every vertex i=1,2,…,N
Iteratively choose pairs of stubs at random and join the m together
Scale-free networks (evolution of networks)
Start with m0 unconnected vertices
Growth: add a new vertex with m stubs at every time step
Preference attachment:
Hierarchical networks
Coexistence of modularity, local clustering, scale-free tology
Generalized random graphs [11]
( )i i / j
j
k k k
14
Complex Networks
15 [12]
Phenomena in Complex Networks: Phase Trans ition
Phase transition: as an external parameter is varied, a change occurs in the macroscopic beh avior of the system under study [10].
Example: Emergence of giant component in gener alized random graphs [13]
Degree distribution : pk
Outgoing degree distribution of neighbors:
With the aid of generating function, [13] derived d istribution of component sizes. Specially, the aver age component size is
Diverges if , and a giant componen t emerges.
For random graphs, a giant component emerges if
( 1) 1 /
k k j
j
q k p
jp2
1 2
2 s k
k k
2 2
k k
( 1) 1 k p N
16
Phenomena in Complex Networks: Synchroniza tion
Synchronization: many natural systems can be d escribed as a collection of oscillators couple d to each other via an interaction matrix and display synchronized behavior [10].
Application: distributed decision through sel f-synchronization [14]
xi(t): state of the system yi: measurement (e.
g. temperature)
gi(yi): local processing unit K: global control loop gain
Ci: local positive coefficient aij: coupling among nodes
h: coupling function w(t): coupling no ise
: propagation delay
1 1
( ) ( ) [ ( ) ( )] ( )
N
i i i i ij j ij i i
j
x t g y KC a h x t x t w t
ij
17
Phenomena in Complex Networks: Synchroniza tion
Form of consensus: when h(x)=x, system achieves synchronize if and only if the directional gra ph is quasi strongly connected (QSC) and
1
1 1
( ) lim ( )
N
i i i i
i
q N N
t
i i i ij ij
i i
c g y x t
c K a
Example of QSC graph [14]
18
Dynamical Processes on Complex Networks
Epidemic spreading
SIR model
S: susceptible, I: infective, R: recovered
Fully mixed model
SIS model
Application in routing/data forwarding in mobile ad hoc networks [15]
Search in networks
Search in power-law random graphs [16]
Random walk
Utilizing high degree nodes
, ,
ds di dr
is is i i
dt dt dt
( )
P k k
3(1 2/ )
s N
s N 2 4/
19
Further Research Topics
Cognition and knowledge construction/representation o f humans
Information theoretical approach to local information
In general, we can model the observing/sensing process as a c hannel, what does the channel capacity mean?
What is relationship between channel capacity and statistical inference?
What are conditions that cooperative information helps (or th ey achieves consensus)?
Example: spectrum sensing in cognitive radio networks
Global informati
on Observed
local informatio Equivalent channel model n
Cooperativ e
informatio n
20
Reference
[1] K. C. Chen, Cognitive radio networks, lecture note.
[2] M. Wang and T. Suda, “The bio-networking architecture: A biologically inspired approach to the design of s calable, adaptive, and survivable/available network application,”
[3] M. Margaliot, “Biomimicry and fuzzy modeling: A match made in heaven,” IEEE Computational Intelligen ce Magazine, Aug. 2008.
[4] M. Dorigo and T. Stutzle, Ant colony optimization, 2004.
[5] S. C. Nicolis, “Communication networks in insect societies,” BIOWIRE, pp. 155-164, 2008.
[6] S. Camazine, J. L. Deneubourg, N. R. Franks, J. Sneyd, G. Theraulaz, and E. Bonabeau, Self-organization in biological systems, 2003.
[7] E. Bonabeau, M. Dorigo, and G. Theraulaz, Swarm intelligence: From natural to artificial systems, 1999.
[8] J. Y. Le Boudec and S. Sarafijanovic, “ An artificial immune system approach to misbehavior detection in m obile ad-hoc networks,” Bio-ADIT, pp. 96-111, Jan. 2004.
[9] M. E. J. Newman, “The structure and function of complex networks,” 2003
[10] A. Barrat, M. Barthelemy, and A. Vespignani, Dynamical processes on complex networks, 2008 [11] C. Gros, Complex and adaptive dynamical systems, 2008.
[12] A-L Barahasi and Z. N. Oltvai, “Network biology: Understanding the cell’s function organization,” Nature Review, Feb. 2004.
21
Reference
[13] M. E. J. Newman, S. H. Strogatz, and D. J. Watts, “Random graphs with arbitrary degree distributions and t heir applications,” Physical Review E., 2001.
[14] S. Barbarossa and G. Scutari, “Bio-inspired sensor network design: Distributed decisions through self-sync hronization,” IEEE Signal Processing Magazine, May 2007.
[15] L. Pelusi, A. Passarella, and M. Conti, “Opportunistic networking: Data forwarding in disconnected mobile ad hoc networks,” IEEE Communications Magazine, Nov. 2006.
[16] L. A. Adamic, R. M. Lukose, A. R. Puniyani, and B. A. Huberman, “Search in power-law networks,” Physi cal Review E., 2001.
22