swarm intelligence course lecture 2
TRANSCRIPT
![Page 1: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/1.jpg)
Chemical Computation
• Slime Mold• Belouzov-Zhabotinski• Reaction diffusion systems• Gray Scott
– http://www.swiss.ai.mit.edu/projects/amorphous/GrayScott/
Good talk onReaction-diffusion
Systems
![Page 2: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/2.jpg)
Brusselator Scheme
A -> XB + X -> Y + waste2X + Y -> 3XX -> waste
Diffusion
Structures maintained by diffusion of Energy
Reactions necessary to maintain structures
A,B have fixed concentrations, rate constant 1.0
Run demo
![Page 3: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/3.jpg)
Self Organization
![Page 4: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/4.jpg)
Emergent Behaviour and Mobile Agents
Tony WhiteBernie Pagurek
![Page 5: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/5.jpg)
Overview
• Why emergent, why not rational?• Characteristics of emergent problem
solving?• SynthECA
– architecture– examples
• Conclusions
![Page 6: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/6.jpg)
Problems of Rationality
• Knowledge of self (and sometimes society)– BDI– Interrap, TouringMachine etc.
• Grounded in symbolic view of world– closed world assumptions– limitations of first order predicate logic
• Frame problem• Tend to be brittle
![Page 7: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/7.jpg)
Characteristics of Emergent Problem Solving
• Aggregation– agents of small “mass”, i.e., simple, reactive
• Non-linearity– interactions are not additive
• Flows– gradients, feedback
• Diversity– multiple agent classes more effective– stochastic behaviour
![Page 8: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/8.jpg)
Emergent Organization: Forces at Play
![Page 9: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/9.jpg)
Cellular Inspiration
![Page 10: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/10.jpg)
SynthECA
• Based upon the Chemical Abstract Machine (CHAM)
• Draws on societies of agents• Problem solving is distributed, emergent
– no global interactions– robust to failure of individual agents
• British Telecom interested in ecological problem solving.
![Page 11: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/11.jpg)
• Swarm agents:– arrive at a node,– sense environment,– undertake local activity,– modify environment,– use sensory input to make migration decision.
SynthECA operation
![Page 12: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/12.jpg)
SynthECA Formalism
A=(E,R,C,MDF,m)• Agents (A) have an architecture consisting
of five components:• emitters (E),• receptors (R),• chemistry (C),• a migration decision function (MDF),• memory (m)
![Page 13: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/13.jpg)
Multi-swarm Architecture
![Page 14: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/14.jpg)
Chemicals
• Chemicals are digitally-encoded using a {1, 0, #} alphabet.
• The # symbol unifies with 1 or 0.• Chemicals have two attributes:
– encoding– concentration
• Chemicals participate in reactions.
![Page 15: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/15.jpg)
• The set of chemical reactions (C) that can operate on sensed and locally stored chemicals.
Chemistry
Chemical interactions
ZWYXZXYX
ZYXYYX
nothingX
+→++→+
→+→+
→ ''
![Page 16: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/16.jpg)
Examples of chemical reactions
Catalytic breakdown of 0110#10#1011 →+
ZWTXZXTX
ZYTYYT
+→++→+
→+→+
TWYXTXYX
TYX
+→++→+
→+
Endothermic reactionsEndothermic reactionsExothermic reactionsExothermic reactions
![Page 17: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/17.jpg)
Migration Decision Function
The MDF is used to determine the next node to visit in the network
pijk (t) = Πp[Tijkp(t) ]-αkp[C(i,j)]-β / Nk(i,j,t)
Nk(i,j,t) = Σj in A(i) Πp [Tijkp(t) ]- αkp[C(i,j)]-β
αkp, β are control parametersNk(i,j,t) is a normalization termA(i) is the set of available egress links
![Page 18: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/18.jpg)
![Page 19: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/19.jpg)
![Page 20: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/20.jpg)
Multi-Swarm scenario
• Distributed Network Management employing delegation [Yemini, 91]
• Three interacting swarms:– connection finding,– connection monitoring,– connection fault diagnosis
A
B
C
D
E
![Page 21: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/21.jpg)
Routing Agent Architecture
• “heavyweight” static agents that make real decisions present at the source node.
• “lightweight” mobile agents traverse network laying down chemical to reinforce path.
AB
C
D
E
F
G
![Page 22: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/22.jpg)
Point to point path
n1
n2n3
n4
n5
n6
n7n8
n9 n11
n10
n12
3
4 4
4 3
33
31
13
3
10
102
7
6 2
![Page 23: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/23.jpg)
Point to multi-point path
n1
n2n3
n4
n5
n6
n7n8
n9 n11
n10
n12
3
4 4
4 3
33
31
13
3
10
102
7
6 2
![Page 24: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/24.jpg)
Protected Path
n1
n2n3
n4
n5
n6
n7n8
n9 n11
n10
n12
3
4 4
4 3
33
31
13
3
10
102
7
6 2
![Page 25: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/25.jpg)
Routing agents
• Agent types:– explorer
• used for route determination– allocator
• allocates resources in network when route emerged– deallocator
• deallocates resources in network when removing connection
![Page 26: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/26.jpg)
Point-2-Point Connections
• For explorer agents:– At each node, they choose path with probability
proportional to f(ce, pe);– explorers visit edges once only (achieved
through use of tabu list);– when destination reached, ants return along the
path explored laying down pheromone trail;– when explorers return a decision is made
regarding path emergence
![Page 27: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/27.jpg)
Path Emergence
• At source node (“nest”):– store paths for previous m explorer agents;– when p% follow same path allocator agent is
sent to allocate bandwidth in network;– explorer agents continue to look for new
(possibly better) paths.• Applies for one or many pt-2-pt
connections: – ants use different, non-reacting pheromones.
![Page 28: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/28.jpg)
Explorer agent algorithm1. Initialize
set t:= 0For every edge (i,j) set an initial value Tij(t) for trail intensity. Place m ants on the source node. [Generate new explorers at freq. ef] ]
2. Set s:= 1 { tabu list index)for k:= 1 to m do
Place starting town of the kthant in tabuk(s).
3. Repeat until dest’n reached:Set s := s + 1for k:=1 to m do
Choose the node j to move towith probability pij
k (t)
Move the kth ant to node j.Update explorer route cost:rk = rk + c(i,j)if (rk > rmax)kill explorerk
Insert town j in tabuk(s).At destination go to 4.
4. While s > 1traverse edge (i,j)T(i,j) = T(i,j) + pe
s := s - 15. At source node do:
if (pathe = pathBuffer * d)create and send allocator
if t > Tmax create and send allocator
Evaporation occurs concurrently with explorationEvaporation occurs concurrently with explorationEvaporation occurs concurrently with exploration
![Page 29: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/29.jpg)
Point-2-Multipoint Connections
• For j destinations, consider as j pt-2-pt connections with:– same pheromone, i.e. all explorers
communicate;– j allocator agents only allocate bandwidth once;– allocator send decision made when % of all j
explorer ants agree on spanning tree.
![Page 30: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/30.jpg)
Routing Function
Transition probability (mdf):pij
k (t) = [Tijk(t) ]-α[C(i,j)]-β / Nk(i,j,t)Nk (i,j,t) = Σj in (S-Tabu(k)) [Tijk(t) ]-α[C(i,j)]-β
α, β are control parameters that determine the sensitivity of the algorithm to link cost and pheromone.
C(i,j) a function that depends upon the type of traffic, the length and utilization of the link.
![Page 31: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/31.jpg)
Allocator agents
• Allocator agents can fail:– bandwidth already allocated by time allocator is
sent;– allocator agent backtracks to source rolling
back resource allocation and decreases pheromone levels;
– decision to re-send allocator made at a later time (a backoff period is observed);
– explorer ants continue to search for routes.
![Page 32: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/32.jpg)
Routing Results
• Shortest paths emerged quickly• Mixed pt-2-pt, pt-2-mpt routes emerged• Routing responds to changes in
environment:– node failure;– link failure;– link cost changes
![Page 33: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/33.jpg)
Parameter Sensitivity
• Bad solutions and stagnation– For high values of α the algorithm enters
stagnation behavior very quickly without finding very good solutions.
• Bad solutions and no stagnation– α too low, insufficient importance associated
with trail.• Good solutions
– α , β in the central area (1,1), (1,2), (1,5), (0.5, 5)
![Page 34: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/34.jpg)
Routing Results
• Link cost functions: C(i,j)Five functions studied experimentally:
• constant• linear• linear threshold• quadratic• server (1/1-u)
At high occupancy (> 50%) server appeared to give the bestresults. At low occupancy (<25%) function relatively unimportant.
constantconstant
linearlinear
thresholdthreshold
quadraticquadratic
serverserver
![Page 35: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/35.jpg)
Fail n3, route is recomputed
![Page 36: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/36.jpg)
Multi-priority routing
• AB (10), CD (11) standard routing agents.• FG (00) is a high priority routing agent.• FG evaporates 10,11 chemicals.
1# → ‘nothing’
AB
C
D
E
F
G
![Page 37: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/37.jpg)
Preferred application routing
• Two (or more) applications should use common path elements:– use same chemical in computing route or,– use wild cards in receptor; e.g. 0#1 to detect
001 and 011.
AB
C
D
E
F
G
![Page 38: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/38.jpg)
Application coalescence/avoidance
• Force applications to go along particular paths.Send agent into network with chemistry:X + Y → Z Agents generating X or Y can also sense Z.
• Force applications to avoid another.Agent uses catalytic conversion:X + Y → Y
![Page 39: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/39.jpg)
Application Oriented RoutingAverage Qos
0
0.02
0.04
0.06
0.08
0.1
1 6 11 16 21 26 31 36 41 46
Generations
Qos
Adapt encodings for pheromones if applications sharing resource have high QoS
![Page 40: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/40.jpg)
Service Dependency Model
• Services (S) depend upon resources (Ri))},{( ii pRS a
For Example:
A PVC depends upon the nodes and links that it passes through. Those links, in turn, depend upon lower level links in the virtual network.
![Page 41: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/41.jpg)
Connection monitoring agents
• Connection’s quality of service monitored.• Changes in QoS cause Connection
monitoring agents to be sent into network, laying down q-chemical.
2
A
B
C
D
•E
![Page 42: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/42.jpg)
Diagnosis
• Diagnosis agent is ‘hill climbing’ in space of q-chemical.
• Very quickly finds high q-chemical concentrations and initiates diagnostic activity.
• An example of distributed diagnosis through constructive chemical interference.
![Page 43: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/43.jpg)
Overlay chemical agents
• Want to separate sensing from reasoning.• Have agents that interact with actual
components, generating chemical messages.• Have “pure” agents that use only chemical
concentrations for movement and problem solving.
![Page 44: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/44.jpg)
Learning
Q s a,( ) Q s a,( ) α r γ maxa' Q s' a',( ) Q s a,( )–+( )+←
Updates to Q values for states are positiveor negative depending upon improvementor degradation in QoS. If QoS improves, Qvalue increases and vice versa.
![Page 45: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/45.jpg)
Connection diagnosis agents
• Examples of netlets• Sense q-chemical concentrations
2A
B
C
D
E)()()(/)()(
tQtNtNtQtp
ijj
ijij
Σ==
Mdf
![Page 46: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/46.jpg)
Problem Agents
• PVC Quality Of Service agent– q-chemical sensing
• Chronic Failure agent– c-chemical sensing
• Overload agent– driven by output of condition sensing agents
![Page 47: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/47.jpg)
Converged Performance vs Agent Number
0.70.91.1
1 2 3 4 5
Number of Agents
![Page 48: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/48.jpg)
Fitness vs Generations
0
0.2
0.4
0.6
0.8
1
1 6 11 16 21 26 31 36 41 46
Generations
Fitn
ess
Fitness vs Generations
0
0.2
0.4
0.6
0.8
1
1 6 11 16 21 26 31 36 41 46
Generations
Fitn
ess
![Page 49: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/49.jpg)
Unreliable components
• Install a broad spectrum evaporator agent (or have it move around network).1##### → ‘nothing’
• Unreliable chemical + congestion chemical used to generate another which is used by an agent to infer that a component is unreliable in situations of congestion.
![Page 50: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/50.jpg)
Symbiont Agent Classes
• Network is unreliable, we may lose agents– need to regenerate agents
• A1=(E1,R1,C1,MDF1,m1),A2=(E2,R2,C2,MDF2,m2), share chemicals in their emitters and receptors; i.e. E1 ∩ R2 ≠Ø and E2 ∩ R1 ≠ Ø.
• A1 generates new A2 if visit frequency too low; A1 dies if visit frequency too high.
![Page 51: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/51.jpg)
Maintaining Agent DensityMaintaining agent density
0
10
20
30
40
50
60
70
80
0 100 200 300 400 500 600 700 800
Time
Sym
bion
t num
ber
Think of dynamics in terms of Lotka-Volterra
![Page 52: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/52.jpg)
Figure 3. Agent # vs Time
0
5
10
15
20
25
0 20 40 60 80 100 120 140 160
Ti me
![Page 53: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/53.jpg)
Maintaining Agent Density
763.088 7762.403 8761.972 7761.252 8760.088 9757.403 8754.252 7753.252 6751.973 5
739.98 4739.721 5
739.02 6735.02 7
734.973 6732.759 5730.631 6729.252 7
728.52 6727.973 7727.047 6
727.02 7726.252 6
726.02 5725.619 4722.252 5
722.02 4714.805 3713.763 4713.612 5713.412 6712.819 7711.079 8
709.96 9708.067 10707.788 11
Stability in the face of unreliability
0
5
10
15
20
25
30
35
0 100 200 300 400 500 600 700 800 900
Time
Num
ber o
f Sym
bion
ts
Think of dynamics in terms of Lotka-Volterra
Agents injected
Agents destroyed
![Page 54: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/54.jpg)
Agent Number vs Time
0
5
10
15
20
25
30
35
40
0 20 40 60 80 100 120 140 160 180
Time
# A
gent
s
Reinjection of flawed
agent
Injection of improved
agent
![Page 55: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/55.jpg)
Introducing a "perfect" agent
0
5
10
15
20
25
30
35
40
0 20 40 60 80 100 120 140 160 180
Time
# A
gent
s
Injection of “perfect”
agent
Injection of improved
agentAddingmoreagents
![Page 56: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/56.jpg)
Resilience in the face of reintroduction of imperfect agents
0
5
10
15
20
25
30
35
40
0 20 40 60 80 100 120 140 160 180
Time
# A
gent
s
System is stableeven with several
competing versions
System is resistantto reintroduction
of obsolete versions
![Page 57: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/57.jpg)
Bread Crumb Algorithm
• Used to configure a service in a network; think of code mobility.
• Inject base class into network and watch it go...
A = {B, C}
C
B
![Page 58: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/58.jpg)
Implementation
“How it works”
![Page 59: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/59.jpg)
Virtual Managed Component
SecurityMethods
Virtual Managed Component Interface (VMCI)
Managed Resources
External parties
Interface toManaged
Resources
ManagementApplets
RecoveryProcedures
ProvisioningProcedures
![Page 60: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/60.jpg)
NC
JVM
DPI- en abledSNMP Agent
VMC withDPI
MCDMCM
SNMPManager
Mobile AgentManager
Sensing Agents
![Page 61: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/61.jpg)
How it works
Mobile Agent Manager
![Page 62: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/62.jpg)
What the press say…Newscientist 24th January, 1998
‘Rama Nune, a senior network designer at MCI, says the ants couldeventually handle most of the day-to-day tasks of running atelephone network--everything from managing the flow of traffic tocalculating everyone's phone bill. "We are trying to exploit theWe are trying to exploit theautonomous intelligence of the ants and the dautonomous intelligence of the ants and the distributed network ofistributed network ofinformation they useinformation they use," says Nune.
MCI also wants to let the ants evolveevolve. All ants implement analgorithm for routing information. But it is possible that the existingalgorithms could be improved upon. One way to find bettersolutions is to let, say, a network management ant "breed" with abilling ant to produce a hybrid. In order to breed, the ants swapsmall segments of their programming code. Each ant will have abuilt-in method for judging how well it does its allotted task, using ameasure known as a fitness function. By killing off unfit ants andallowing the fittest to carry on breeding, it may be possible toproduce ants that do their jobs better than any human-designed ant.’
![Page 63: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/63.jpg)
Futuristic?
• Active networks are being researched at:– M.I.T.– U. Penn.– CMU– Georgia Tech.
• Management by delegation [Yemini, 91] considered an essential design criterion for next generation network management systems.
![Page 64: Swarm Intelligence Course Lecture 2](https://reader033.vdocuments.us/reader033/viewer/2022060211/62935b3e23a13f2e744ba60a/html5/thumbnails/64.jpg)
Where next?
• Build a complex web of interacting agents, assess stability.
• Formulate a model based on reaction kinetics in order to provide analytical framework.
Good thesis workHere!