+ All Categories
Home > Documents > aco-03-04-2013

aco-03-04-2013

Date post: 17-Dec-2015
Category:
Upload: rahul-bhardwaj
View: 213 times
Download: 1 times
Share this document with a friend
Description:
gf
Popular Tags:
70
Ant Colony Optimization: an introduction Daniel Chivilikhin 03.04.2013
Transcript
  • Ant Colony Optimization: an introductionDaniel Chivilikhin03.04.2013

  • OutlineBiological inspiration of ACOSolving NP-hard combinatorial problemsThe ACO metaheuristicACO for the Traveling Salesman Problem

  • OutlineBiological inspiration of ACOSolving NP-hard combinatorial problemsThe ACO metaheuristicACO for the Traveling Salesman Problem

  • Biological inspiration: from real to artificial ants

  • Ant coloniesDistributed systems of social insectsConsist of simple individualsColony intelligence >> Individual intelligence

  • Ant CooperationStigmergy indirect communication between individuals (ants)Driven by environment modifications

  • Denebourgs double bridge experimentsStudied Argentine ants I. humilisDouble bridge from ants to food source

  • Double bridge experiments: equal lengths (1)

  • Double bridge experiments: equal lengths (2)Run for a number of trialsAnts choose each branch ~ same number of trials

  • Double bridge experiments: different lengths (2)

  • Double bridge experiments: different lengths (2) The majority of ants follow the short path

  • OutlineBiological inspiration of ACOSolving NP-hard combinatorial problemsThe ACO metaheuristicACO for the Traveling Salesman Problem

  • Solving NP-hard combinatorialproblems

  • Combinatorial optimizationFind values of discrete variablesOptimizing a given objective function

  • Combinatorial optimization = (S, f, ) problem instanceS set of candidate solutionsf objective function set of constraints set of feasible solutions (with respect to )Find globally optimal feasible solution s*

  • NP-hard combinatorial problemsCannot be exactly solved in polynomial timeApproximate methods generate near-optimal solutions in reasonable timeNo formal theoretical guaranteesApproximate methods = heuristics

  • Approximate methodsConstructive algorithmsLocal search

  • Constructive algorithmsAdd components to solution incrementallyExample greedy heuristics: add solution component with best heuristic estimate

  • Local searchExplore neighborhoods of complete solutionsImprove current solution by local changesfirst improvementbest improvement

  • What is a metaheuristic?A set of algorithmic conceptsCan be used to define heuristic methodsApplicable to a wide set of problems

  • Examples of metaheuristicsSimulated annealingTabu searchIterated local searchEvolutionary computationAnt colony optimizationParticle swarm optimizationetc.

  • OutlineBiological inspiration of ACOSolving NP-hard combinatorial problemsThe ACO metaheuristicACO for the Traveling Salesman Problem

  • The ACO metaheuristic

  • ACO metaheuristicA colony of artificial ants cooperate in finding good solutionsEach ant simple agentAnts communicate indirectly using stigmergy

  • Combinatorial optimization problem mapping (1)Combinatorial problem (S, f, (t))(t) time-dependent constraintsExample dynamic problemsGoal find globally optimal feasible solution s*Minimization problemMapped on another problem

  • Combinatorial optimization problem mapping (2)C = {c1, c2, , cNc} finite set of componentsStates of the problem:X = {x = , |x| < n < +}Set of candidate solutions:

  • Combinatorial optimization problem mapping (3)Set of feasible states:

    We can complete into a solution satisfying (t)Non-empty set of optimal solutions:

  • Combinatorial optimization problem mapping (4) X states S candidate solutions feasible states S* optimal solutions

    S*

  • Combinatorial optimization problem mapping (5)Cost g(s, t) for each In most cases g(s, t) f(s, t) GC = (C, L) completely connected graphC set of componentsL edges fully connecting the components (connections)GC construction graph

  • Combinatorial optimization problem mapping (last )Artificial ants build solutions by performing randomized walks on GC(C, L)

  • Construction graphEach component ci or connection lij have associated:heuristic informationpheromone trail

  • Heuristic informationA priori information about the problemDoes not depend on the antsOn components ci iOn connections lij ijMeaning: cost of adding a component to the current solution

  • Pheromone trailLong-term memory about the entire search processOn components ci iOn connections lij ijUpdated by the ants

  • Artificial ant (1)Stochastic constructive procedureBuilds solutions by moving on GCHas finite memory for:Implementing constraints (t)Evaluating solutionsMemorizing its path

  • Artificial ant (2)Has a start state xHas termination conditions ekFrom state xr moves to a node from the neighborhood Nk(xr)Stops if some ek are satisfied

  • Artificial ant (3)Selects a move with a probabilistic rule depending on:Pheromone trails and heuristic information of neighbor components and connectionsMemoryConstraints

  • Artificial ant (4)Can update pheromone on visited components (nodes)and connections (edges)Ants act:ConcurrentlyIndependently

  • The ACO metaheuristicWhile not doStop():ConstructAntSolutions() UpdatePheromones() DaemonActions()

  • ConstructAntSolutionsA colony of ants build a set of solutionsSolutions are evaluated using the objective function

  • UpdatePheromonesTwo opposite mechanisms:Pheromone depositPheromone evaporation

  • UpdatePheromones: pheromone depositAnts increase pheromone values on visited components and/or connectionsIncreases probability to select visited components later

  • UpdatePheromones: pheromone evaporationDecrease pheromone trails on all components/connections by a same valueForgetting avoid rapid convergence to suboptimal solutions

  • DaemonActionsOptional centralized actions, e.g.:Local optimizationAnt elitism (details later)

  • ACO applicationsTraveling salesmanQuadratic assignmentGraph coloringMultiple knapsackSet coveringMaximum cliqueBin packing

  • OutlineBiological inspiration of ACOSolving NP-hard combinatorial problemsThe ACO metaheuristicACO for the Traveling Salesman Problem

  • ACO for the Traveling Salesman Problem

  • Traveling salesman problemN set of nodes (cities), |N| = nA set of arcs, fully connecting NWeighted graph G = (N, A)Each arc has a weight dij distanceProblem:Find minimum length Hamiltonian circuit

  • TSP: construction graphIdentical to the problem graphC = NL = Astates = set of all possible partial tours

  • TSP: constraintsAll cities have to be visitedEach city only onceEnforcing allow ants only to go to new nodes

  • TSP: pheromone trailsDesirability of visiting city j directly after i

  • TSP: heuristic informationij = 1 / dijUsed in most ACO for TSP

  • TSP: solution constructionSelect random start cityAdd unvisited cities iterativelyUntil a complete tour is built

  • ACO algorithms for TSPAnt SystemElitist Ant SystemRank-based Ant SystemAnt Colony SystemMAX-MIN Ant System

  • Ant System: Pheromone initializationPheromone initializationij = m / Cnn, where:m number of antsCnn path length of nearest-neighbor algorithm

  • Ant System: Tour constructionAnt k is located in city i is the neighborhood of city iProbability to go to city :

  • Tour construction: comprehension = 0 greedy algorithm = 0 only pheromone is at workquickly leads to stagnation

  • Ant System: update pheromone trails evaporationEvaporation for all connections(i, j) L:ij (1 ) ij, [0, 1] evaporation ratePrevents convergence to suboptimal solutions

  • Ant System: update pheromone trails depositTk path of ant kCk length of path TkAnts deposit pheromone on visited arcs:

  • Elitist Ant SystemBest-so-far ant deposits pheromone on each iteration:

  • Rank-based Ant SystemRank all antsEach ant deposits amounts of pheromone proportional to its rank

  • MAX-MIN Ant SystemOnly iteration-best or best-so-far ant deposits pheromonePheromone trails are limited to the interval [min, max]

  • Ant Colony SystemDiffers from Ant System in three points:More aggressive tour construction ruleOnly best ant evaporates and deposits pheromoneLocal pheromone update

  • Ant Colony SystemTour Construction

    Local pheromone update:ij (1 )ij + 0,

  • Comparing Ant System variants

  • State of the art in TSPCONCORDE http://www.tsp.gatech.edu/concorde.html Solved an instance of 85900 citiesComputation took 286-2719 CPU days!

  • Current ACO research activityNew applicationsTheoretical proofs

  • Further readingM. Dorigo, T. Sttzle. Ant Colony Optimization. MIT Press, 2004.http://iridia.ulb.ac.be/~mdorigo/ACO/

  • Next timeSome proofs of ACO convergence

  • Thank you!

    Any questions?

    This presentation is available at:http://rain.ifmo.ru/~chivdan/presentations/Daniel Chivilikhin [mailto: [email protected]]

  • Used resourceshttp://teamaltman.com/wp-content/uploads/2011/06/Uncertainty-Ant-Apple-1024x1024.jpg http://myrealestatecoach.files.wordpress.com/2012/04/ant.jpghttp://ars.els-cdn.com/content/image/1-s2.0-S1568494613000264-gr3.jpghttp://www.theorie.physik.uni-goettingen.de/forschung/ha/talks/stuetzle.pdf http://www.buyingandsellingwebsites.com/wp-content/uploads/2011/12/Ant-150.jpghttp://moodle2.gilbertschools.net/moodle/file.php/1040/Event-100_Days/Ant_Hormiga.gifhttp://4.bp.blogspot.com/_SPAe2p8Y-kg/TK3hQ8BUMtI/AAAAAAAAAHA/p75K-GcT_oo/s1600/ant+vision.GIFhttp://1.bp.blogspot.com/-tXKJQ4nSqOY/UQZ3vabiy_I/AAAAAAAAAjA/IK8jtlhElqk/s1600/16590492-illustration-of-an-ant-on-a-white-background.jpg


Recommended