ant colony optimization overview

Upload: mohammed-lubbad

Post on 06-Apr-2018

233 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/2/2019 Ant Colony Optimization Overview

    1/31

    An Overview of Swarm

    Intelligence and Ant ColonyOptimization Heuristics

    Philipp A. Djang Ph.D.

    Army Research Labs

    "Go to the ant, thou sluggard; consider her ways, and be wise:Which having no guide, overseer, or ruler,

    Provideth her meat in the summer,and gathereth her food in the harvest"

    (Proverbs vi 6-8)

  • 8/2/2019 Ant Colony Optimization Overview

    2/31

    Overview

    Swarm Intelligence

    Ant Colony Algorithm

    Solving a Traveling SalespersonProblem

    Other Examples

    References

  • 8/2/2019 Ant Colony Optimization Overview

    3/31

    Swarm Intelligence Swarm Intelligence (SI) is the property of a

    system whereby the collective behaviors of(unsophisticated) agents interacting locallywith their environment cause coherentfunctional global patterns to emerge.

    SI provides a basis with which it is possible toexplore collective (or distributed) problemsolving without centralized control or theprovision of a global model.

    Leverage the power of complex adaptivesystems to solve difficult non-linear stochasticproblems

  • 8/2/2019 Ant Colony Optimization Overview

    4/31

    Swarm Intelligence

    Characteristics of a swarm:

    Distributed, no central control or data

    source; Limited communication

    No (explicit) model of the environment;

    Perception of environment (sensing) Ability to react to environment changes.

  • 8/2/2019 Ant Colony Optimization Overview

    5/31

    Swarm Intelligence

    Social interactions (locally sharedknowledge) provides the basis for

    unguided problem solving The efficiency of the effort is related to

    but not dependent upon the degree or

    connectedness of the network and thenumber of interacting agents

  • 8/2/2019 Ant Colony Optimization Overview

    6/31

    Swarm Intelligence

    Robust exemplars of problem-solving inNature Survival in stochastic hostile environment

    Social interaction creates complexbehaviors

    Behaviors modified by dynamicenvironment.

    Emergent behavior observed in: Bacteria, immune system, ants, birds

    And other social animals

  • 8/2/2019 Ant Colony Optimization Overview

    7/31

    Ants Swarm Intelligence

    Example Franks observed Lasius Nigerants,

    regulation of 1 degree Celsius range;

    forming bridges;

    raiding specific areas for food; building and protecting nest;

    sorting brood and food items;

    cooperating in carrying large items;

    emigration of a colony;

    finding shortest route from nest to food source; preferentially exploiting the richest food source available.

    Without Any Central Leadership or Control

  • 8/2/2019 Ant Colony Optimization Overview

    8/31

    Ant Colony Optimization:

    Introduction First proposed by M. Dorigo, 1992

    Heuristic optimization method inspired by

    biological systems Multi-agent approach for solving difficult

    combinatorial optimization problems

    Traveling Salesman, vehicle routing, sequential

    ordering, graph coloring, routing incommunications networks

    Has become new and fruitful research area

  • 8/2/2019 Ant Colony Optimization Overview

    9/31

    Ant Colony Algorithms Algorithm was inspired by observation

    of real ant colonies.

    Ants are essentially blind, deaf and

    dumb. Ants are social creatures behavior

    directed to survival of colony

    Q: how can ants find the short path tofood sources?

    Ants deposit pheromoneson groundthat form a trail. The trail attracts otherants.

  • 8/2/2019 Ant Colony Optimization Overview

    10/31

    Ant Colony Algorithms

    Ant behavior is a kind of stochasticdistributed optimization behavior.

    Although one ant is capable of buildinga solution, it is the behavior of anensemble of ants that exhibits theshortest path behavior.

    The behavior is induced by indirectcommunication (pheromone paths)without central control.

  • 8/2/2019 Ant Colony Optimization Overview

    11/31

    Ant Colony Algorithms Ants do not know the global structure of

    the problem - discoverthe network

    Limited ability to sense localenvironment - can only see adjacentnodes of immediate neighborhood.

    Each ant chooses an action based onvariableprobability

    random choice

    pheromone mediated

  • 8/2/2019 Ant Colony Optimization Overview

    12/31

    Ant Colony Algorithms

    Each ant collects information aboutlocal environment; acts concurrently

    and independentlyNo direct communication: stigmergy

    paradigm governs information exchange

    Incremental constructive approach tobuilding solutions

    High quality solutions emerge via globalcooperation.

  • 8/2/2019 Ant Colony Optimization Overview

    13/31

    Stigmergy

    Indirect communication via interactionwith environment [Gass, 59, Wilson,75] Sematonic stigmergy

    action of agent directly related to problemsolving and affects behavior of other agents.

    Sign-based stigmergy action of agent affects environment not directly

    related to problem solving activity.

  • 8/2/2019 Ant Colony Optimization Overview

    14/31

    Species lay pheromone trails traveling fromnest, to nest or possibly in both directions.

    Pheromones evaporate. Pheromones accumulate with multiple ants

    using path.

    Pheromone Trails

    Foodsource

    Nest

  • 8/2/2019 Ant Colony Optimization Overview

    15/31

    Pheromone Trails Example

    D

    E

    H C

    A

    B

    d=0.5

    d=0.5

    d=1.0

    d=1.0

    15ants

    E

    D

    H C

    A

    B

    30 ants

    15

    ants15

    ants

    15ants

    30

    ants

    T =

    0

    E

    H

    D

    C

    A

    B

    30 ants

    10

    ants20

    ants

    20

    ants

    10

    ants

    30

    ants

    T =

    1

  • 8/2/2019 Ant Colony Optimization Overview

    16/31

    Ant Colony Algorithms Pheromone mediated following

    behavior induces the emergence ofshortest paths.

    Probability of choosing a branch of apath at a certain time depends on thetotal amount of pheromone on thebranch.

    The choice is proportional to thenumber of ants that have used thebranches.

  • 8/2/2019 Ant Colony Optimization Overview

    17/31

    Ant Colony Algorithms

    Let um and lm be the number of ants thathave used the upper and lower

    branches. The probability Pu(m) with which the

    (m+1)th ant chooses the upper branch

    is:

    )()(

    )()(

    klku

    kuP

    mm

    mm

    hh

    h

    u

  • 8/2/2019 Ant Colony Optimization Overview

    18/31

    Traveling Salesperson Problem

    Famous NP-Hard Optimization Problem

    Given a fully connected, symmetric

    G(V,E) with known edge costs, find theminimum cost tour.

    Artificial ants move from vertex to vertex

    to order to find the minimum cost tourusing only pheromone mediated trails.

  • 8/2/2019 Ant Colony Optimization Overview

    19/31

    Traveling Salesperson Problem

    The three main ideas that this antcolony algorithm has adopted from real

    ant colonies are: The ants have a probabilistic preference

    for paths with high pheromone value

    Shorter paths tend to have a higher rate of

    growth in pheromone value It uses an indirect communication system

    through pheromone in edges

  • 8/2/2019 Ant Colony Optimization Overview

    20/31

    Traveling Salesperson Problem

    Ants select the next vertex based on aweighted probability function based on twofactors: The number of edges and the associated cost The trail (pheromone) left behind by other ant

    agents.

    Each agent modifies the environment in twodifferent ways :

    Local trail updating: As the ant moves betweencities it updates the amount of pheromone on theedge

    Global trail updating: When all ants havecompleted a tour the ant that found the shortestroute updates the edges in its path

  • 8/2/2019 Ant Colony Optimization Overview

    21/31

    Traveling Salesperson Problem

    Local Updating is used to avoid verystrong pheromone edges and hence

    increase exploration (and hopefullyavoid locally optimal solutions).

    The Global Updating function gives the

    shortest path higher reinforcement byincreasing the amount of pheromone onthe edges of the shortest path.

  • 8/2/2019 Ant Colony Optimization Overview

    22/31

    Empirical Results

    Compared Ant Colony Algorithm tostandard algorithms and meta-heuristic

    algorithms on Oliver 30 a 30 city TSP Standard: 2-Opt, Lin-Kernighan,

    Meta-Heuristics: Tabu Search andSimulated Annealing

    Conducted 10 replications of eachalgorithm and provided averaged results

  • 8/2/2019 Ant Colony Optimization Overview

    23/31

    Comparison to Standard

    Algorithms Examined Solution

    Quality not speed;

    in general, standardalgorithms weresignificantly faster.

    Best ACO solution -

    420

    2-Opt L-K

    NearNeighbor

    437 421

    Far Insert 421 420

    Near Insert 492 420

    Space Fill 431 421

    Sweep 426 421

    Random 663 421

  • 8/2/2019 Ant Colony Optimization Overview

    24/31

    Comparison to Meta-Heuristic

    Algorithms Meta-Heuristics are algorithms that can be

    applied to a variety of problems with a minimumof customization.

    Comparing ACO to other Meta-heuristicsprovides a fair market comparison (vice TSPspecific algorithms).

    Best Mean Std Dev

    ACO 420 420.4 1.3

    Tabu 420 420.6 1.5

    SA 422 459.8 25.1

  • 8/2/2019 Ant Colony Optimization Overview

    25/31

    Other Application Areas

    Scheduling : Scheduling is awidespread problem of practical

    importance. Paul Forsyth & Anthony Wren,

    University of Leeds Computer Science

    department developed a bus driverscheduling application using ant colonyconcepts.

  • 8/2/2019 Ant Colony Optimization Overview

    26/31

    Other Application Areas Telecommunication Networks : Network

    routingrefers to the activity of creating,maintaining and using routing tables (one foreach node in the network) to determinewhere to direct an incoming data stream sothat it can continue its travel through thenetwork.

    In telecommunications, this is an extremelydifficult problem because of the constantchanges in network traffic load. The AntColony algorithm provides adaptiveadvantages that can adjust to traffic load.

  • 8/2/2019 Ant Colony Optimization Overview

    27/31

    Other Application Areas

    Vehicle Routing Problem: The VRP issimilar to the TSP, but is complicated by

    multiple vehicles, vehicle capacity, pick-up and drop off points (which candictate vehicle packing and scheduling).

    Bernd Mullenheimer, Richard Hartl andChristine Strauss developed an AntColony algorithm for solving the VRP

  • 8/2/2019 Ant Colony Optimization Overview

    28/31

    Ant Colony Algorithms: Summary

    Ant Colony Algorithms mimic Real Ants

    Colony of cooperating individuals

    Simulated Pheromone Trail and Stigmergy

    Shortest path searching with local moves

    Stochastic and myopic state transitionpolicy

    Artificial ants:Discrete state transitions

    Pheromones based on solution quality

    Pheromone laying is problem dependent

  • 8/2/2019 Ant Colony Optimization Overview

    29/31

    Interesting Reading

    Alexandrov D., Kochetov Y. Behavior of the Ant ColonyAlgorithm for the Set Covering Problem, Proc. of Symposium.on Operations. Research., Springer Verlag, 2000

    On the MAX/MIN Ant system, Thomas Sttzle, 2001.

    Hybrid Ant System for the Sequential Ordering Problems, LucaGambardella, 2002.

    Parallelization Strategies for Ant Colony Optimization byThomas Sttzle. In Proceedings of PPSN-V, Amsterdam,Springer Verlag, LNCS 1998

    Improvements on the Ant System: Introducing the MAX-MIN Ant

    System by Thomas Sttzle. Proceedings of Artificial Neural Netsand Genetic Algorithms 1997

    The Ant System Applied to the Quadratic Assignment Problemby Maniezzo, Colorni and Dorigo. Tech. Rep. IRIDIA/94-28,Universit Libre de Bruxelles 1994

    http://aida.intellektik.informatik.th-darmstadt.de/~tom/tsp.htmlhttp://www.idsia.ch/~luca/has-sop.htmlhttp://aida.intellektik.informatik.th-darmstadt.de/~tom/publications/PPSN-V.ps.gzhttp://aida.intellektik.informatik.th-darmstadt.de/~tom/publications/ICANNGA97.ps.gzhttp://aida.intellektik.informatik.th-darmstadt.de/~tom/publications/ICANNGA97.ps.gzftp://iridia.ulb.ac.be/pub/dorigo/tec.reps/TR.03-ANT-QAP.ps.gzftp://iridia.ulb.ac.be/pub/dorigo/tec.reps/TR.03-ANT-QAP.ps.gzhttp://aida.intellektik.informatik.th-darmstadt.de/~tom/publications/ICANNGA97.ps.gzhttp://aida.intellektik.informatik.th-darmstadt.de/~tom/publications/ICANNGA97.ps.gzhttp://aida.intellektik.informatik.th-darmstadt.de/~tom/publications/ICANNGA97.ps.gzhttp://aida.intellektik.informatik.th-darmstadt.de/~tom/publications/ICANNGA97.ps.gzhttp://aida.intellektik.informatik.th-darmstadt.de/~tom/publications/PPSN-V.ps.gzhttp://www.idsia.ch/~luca/has-sop.htmlhttp://aida.intellektik.informatik.th-darmstadt.de/~tom/tsp.html
  • 8/2/2019 Ant Colony Optimization Overview

    30/31

    Interesting Reading

    Dorigo, M., Maniezzo, V., Colorni, A., The AntSystem: Optimization by a Colony of CooperatingAgents, IEEE Transactions on Systems, Man andCybernetics-Part B, v26,n1, 1996

    Rafael S. Parpinelli and Heitor S. Lopes and Alex A.Freitas, An Ant Colony Based System for DataMining: Applications to Medical Data, Proceedings ofthe Genetic and Evolutionary ComputationConference ({GECCO}-2001)

    Nicolas Monmarch, Mohamed Slimane, GillesVenturini,AntClass: discovery of clusters in numericdata by an hybridization of an ant colony with thekmeans algorithm, 1999

  • 8/2/2019 Ant Colony Optimization Overview

    31/31

    On-Line Resources

    http://www.swarm.org/

    http://www.swarm-bots.org/

    http://dsp.jpl.nasa.gov/members/payman/swarm/

    http://www.engr.iupui.edu/~shi/pso.html

    http://iridia.ulb.ac.be/~mdorigo/ACO/ACO.html

    http://www.cs.technion.ac.il/~wagner/

    http://ants.gsfc.nasa.gov/