Transcript Seminar Mishra, Ayush

Swarm Intelligence
Swarms





Natural phenomena as inspiration
A flock of birds sweeps across the Sky.
How do ants collectively forage for food?
How does a school of fish swims, turns together?
They are so ordered.





What made them to be so ordered?
There is no centralized controller
But they exhibit complex global behavior.
Individuals follow simple rules to interact with
neighbors .
Rules followed by birds



collision avoidance
velocity matching
Flock Centering
Swarm Intelligence-Definition

“Swarm intelligence (SI) is artificial intelligence
based on the collective behavior of decentralized,
self-organized systems”
Characteristics of Swarms




Composed of many individuals
Individuals are homogeneous
Local interaction based on simple rules
Self-organization
Overview





Ant colony optimization
TSP
Bees Algorithms
Comparison between bees and ants
Conclusions
Ant Colony Optimization



The way ants find their food in shortest path is
interesting.
Ants secrete pheromones to remember their path.
These pheromones evaporate with time.
Ant Colony Optimization..



Whenever an ant finds food , it marks its return
journey with pheromones.
Pheromones evaporate faster on longer paths.
Shorter paths serve as the way to food for most of
the other ants.
Ant Colony Optimization


The shorter path will be reinforced by the
pheromones further.
Finally , the ants arrive at the shortest path.
Optimizations of SI




Swarms have the ability to solve problems
Ant Colony Optimization (ACO) , a meta-heuristic
ACO can be used to solve hard problems like TSP,
Quadratic Assignment Problem(QAP)
We discuss ACO meta-heuristic for TSP
ACO-TSP



Given a graph with n nodes, should give the
shortest Hamiltonian cycle
m ants traverse the graph
Each ant starts at a random node
Transitions


Ants leave pheromone trails when they make a
transition
Trails are used in prioritizing transition
Transitions




Suppose ant k is at u.
Nk(u) be the nodes not visited by k
Tuv be the pheromone trail of edge (u,v)
k jumps from u to a node v in Nk(u) with
probability
puv(k) = Tuv ( 1/ d(u,v))
Iteration of AOC-TSP




m ants are started at random nodes
They traverse the graph prioritized on trails and
edge-weights
An iteration ends when all the ants visit all nodes
After each iteration, pheromone trails are updated.
Updating Pheromone trails

New trail should have two components




Old trail left after evaporation and
Trails added by ants traversing the edge during the
iteration
T'uv = (1-p) Tuv + ChangeIn(Tuv)
Solution gets better and better as the number of
iterations increase
Performance of TSP with ACO heuristic


Performs better than state-of-the-art TSP
algorithms for small (50-100) of nodes
The main point to appreciate is that Swarms give
us new algorithms for optimization
Bee Algorithm
Bees Foraging

Recruitment Behaviour :





Waggle Dancing
series of alternating left and right loops
Direction of dancing
Duration of dancing
Navigation Behaviour :


Path vector represents knowledge representation of
path by inspect
Construction of PI.
Algorithm

It has two steps :




ManageBeesActivity()
CalculateVectors()
ManageBeesActivity: It handles agents activities
based on their internal state. That is it decides
action it has to take depending on the knowledge it
has.
CalculateVectors : It is used for administrative
purposes and calculates PI vectors for the agents.
Uses of Bee Algorithm




Training neural networks for pattern recognition
Forming manufacturing cells.
Scheduling jobs for a production machine.
Data clustering
Comparisons




Ants use pheromones for back tracking route to
food source.
Bees instead use Path Integration. Bees are able to
compute their present location from past trajectory
continuously.
So bees can return to home through direct route
instead of back tracking their original route.
Does path emerge faster in this algorithm.
Results

Experiments with different test cases on these
algorithms show that.



Bees algorithm is more efficient when finding and
collecting food, that is it takes less number of steps.
Bees algorithm is more scalable it requires less
computation time to complete task.
Bees algorithm is less adaptive than ACO.
Applications of SI


In Movies : Graphics in movies like Lord of the
Rings trilogy, Troy.
Unmanned underwater vehicles(UUV):



Groups of UUVs used as security units
Only local maps at each UUV
Joint detection of and attack over enemy vessels by coordinating within the group of UUVs
More Applications

Swarmcasting:




For fast downloads in a peer-to-peer file-sharing
network
Fragments of a file are downloaded from different
hosts in the network, parallelly.
AntNet : a routing algorithm developed on the
framework of Ant Colony Optimization
BeeHive : another routing algorithm modelled on
the communicative behaviour of honey bees
A Philosophical issue




Individual agents in the group seem to have no
intelligence but the group as a whole displays
some intelligence
In terms of intelligence, whole is not equal to sum
of parts?
Where does the intelligence of the group come
from ?
Answer : Rules followed by individual agents
Conclusion




SI provides heuristics to solve difficult
optimization problems.
Has wide variety of applications.
Basic philosophy of Swarm Intelligence : Observe
the behaviour of social animals and try to mimic
those animals on computer systems.
Basic theme of Natural Computing: Observe
nature, mimic nature.
Bibliography




A Bee Algorithm for Multi-Agents SystemLemmens ,Steven . Karl Tuyls, Ann Nowe -2007
Swarm Intelligence – Literature Overview, Yang
Liu , Kevin M. Passino. 2000.
www.wikipedia.org
The ACO metaheuristic: Algorithms, Applications,
and Advances. Marco Dorigo and Thomas StutzleHandbook of metaheuristics, 2002.