A Graph-Based Evolutionary Algorithm: Genetic Network Programming(GNP) and Its Extension Using Reinforcement Learning

Size: px
Start display at page:

Download "A Graph-Based Evolutionary Algorithm: Genetic Network Programming(GNP) and Its Extension Using Reinforcement Learning"

Transcription

1 Graph-Based Evolutionary lgorithm: Genetic Network Programming(GNP) and Its Extension Using Reinforcement Learning Shingo Mabu Graduate School of Information, Production and Systems, Waseda University, Hibikino 2-7 Wakamatsu-ku, Kitakyushu, Fukuoka, , Japan Kotaro Hirasawa Graduate School of Information, Production and Systems, Waseda University, Hibikino 2-7, Wakamatsu-ku, Kitakyushu, Fukuoka, , Japan Jinglu Hu Graduate School of Information, Production and Systems, Waseda University, Hibikino 2-7, Wakamatsu-ku, Kitakyushu, Fukuoka, , Japan bstract his paper proposes a graph-based evolutionary algorithm called Genetic Network Programming(GNP). Our goal is to develop GNP, which can deal with dynamic environments efficiently and effectively, based on the distinguished expression ability of the graph(network) structure. he characteristics of GNP are as follows. 1) GNP programs are composed of a number of nodes which execute simple judgment/processing, and these nodes are connected by directed links to each other. 2) hegraphstructureenablesgnptore-usenodes,thusthestructurecanbeverycompact. 3) he node transition of GNP is executed according to its node connections without any terminal nodes, thus the past history of the node transition affects the current node to be used and this characteristic works as an implicit memory function. hese structural characteristics are useful for dealing with dynamic environments. Furthermore, we propose an extended algorithm, GNP with Reinforcement Learning(GNP- RL) which combines evolution and reinforcement learning in order to create effective graph structures and obtain better results in dynamic environments. In this paper, we applied GNP to the problem of determining agents behavior to evaluate its effectiveness. ileworld was used as the simulation environment. he results show some advantages for GNP over conventional methods. Keywords Evolutionary computation, graph structure, reinforcement learning, agent, tileworld. 1 Introduction large number of studies have been conducted on evolutionary optimization techniques. Genetic lgorithm(g)(holland, 1975), Genetic Programming(GP)(Koza, 1992, 1994) and Evolutionary Programming(EP)(Fogel et al., 1966; Fogel, 1994) are typical evolutionary algorithms. G evolves strings and is mainly applied to optimizationproblems. GPwasdevisedlaterinordertoexpandtheexpressionabilityofG by using tree structures. EP is a graph structural system creating finite state machines by evolution. In this paper, a new graph-based evolutionary algorithm named Genetic c 27 by the Massachusetts Institute of echnology Evolutionary Computation 15(3):

2 S.Mabu,K.HirasawaandJ.Hu Network Programming(GNP)(Katagiri et al., 2, 21; Hirasawa et al., 21; Mabu etal.,22,24)isdescribed. OuraimindevelopingGNPistodealwithdynamic environments efficiently and effectively by using the higher expression ability of graph structure, and the inherently equipped functions in it. he distinguishing functions of the GNP structure are directed graph expression, reusability of nodes, and implicit memory function. he directed graph expression can realize some repetitive processes, and it can be effective because it works like the utomatically Defined Functions(DFs) in GP. he node transition of GNP starts from astartnodeandcontinuesbasedonthenodeconnections,thusitcanbesaidthatan agent s 1 actionsinthepastareimplicitlymemorizedinthenetworkflow. In addition, we propose an extended algorithm, GNP with Reinforcement Learning(GNP-RL) which combines evolution and reinforcement learning in order to create effective graph structures and obtain better results in dynamic environments. he distinguishing functions of GNP-RL are the combinations of offline and online learning, and diversified and intensified search. lthough we have already proposed a method, online learning of GNP(Mabu et al., 23), which uses only reinforcement learning to selectthenodeconnections,thismethodhasaprobleminthattheqtablebecomesvery large, and the calculation time and the occupation of the memory also become large. hus, we propose a method using both evolution and reinforcement learning in this paper. Evolutionary algorithms are superior in terms of wide space search ability because they continue to evolve various individuals and select better ones(offline learning), while RL can learn incrementally, based on rewards obtained during task execution (online learning). herefore, the combination of evolution and RL can cooperatively make good graph structures. In fact, the proposed evolutionary algorithm(diversified search) makes rough graph structures by selecting some important functions among many kinds of functions and connects them based on fitness values after task execution, thus the Q table becomes quite compact. nd then RL(intensified search) selects the best function during task execution, i.e., determines an appropriate node transition. his paper is organized as follows. Section 2 provides the related work and the comparisons with GNP. In Section 3, the details of GNP and GNP-RL are described. Section 4 explains the ileworld problem, available function sets, and shows the simulation results. Section 5 discusses future work and remaining problems. Section 6 is devoted to conclusions. 2 Related Work and Comparisons GPtreecanbeusedasadecisiontreewhenallfunctionnodesare if-thentype functions and all terminal nodes are concrete action functions. In this case, a tree is executedfromarootnodetoacertainterminalnodeineachstep,sothebehaviorsof agents are determined mainly by the current information. On the other hand, since GNPhasanimplicitmemoryfunction,GNPcandetermineanactionbynotonlythe current, but also the past information. he most important problem for GP is the bloat of the tree. he increase in depth causes an exponential enlargement of the search space, the occupation of large amounts of memory, and an increase in calculation time. Constrainingthedepthofthetreeisoneofthewaystoovercomethebloatproblem. SincethegraphstructureofGNPhasanimplicitmemoryfunctionandtheabilityto re-use nodes, GNP is expected to use necessary nodes repeatedly and create compact 1 n agent is a computer system that is situated in some environments, and it is capable of autonomous action in these environments in order to meet its design objectives(weiss, 1999). In this paper, the autonomous action is determined by GNP. 37 Evolutionary Computation Volume 15, Number 3

3 Genetic Network Programming and Its Extension structures. We will have a discussion on the program size in the simulation section. Evolutionary Programming(EP) is a graph structural system used for the automatic synthesis of finite state machines(fsms). For example, FSM programs are evolved in the iterated prisoner s dilemma game(fogel, 1994; ngeline, 1994) and the ant problem(ngeline and Pollack, 1993). However, there are some essential differences between EP and GNP. Generally FSM must define their transition rules for all combinations of states and possible inputs, thus the FSM program will become large andcomplexwhenthenumberofstatesandinputsislarge. InGNP,thenodesare connected by necessity, so it is possible that only the essential inputs obtained in the currentsituationareusedinthenetworkflow.saresult,thegraphstructureofgnp can be quite compact. PDO(eller and Veloso, 1995, 1996; eller, 1996) is also a graph-based algorithm, butitsfundamentalconceptisdifferentfromthatofgnp.eachnodeinpdoprogramshastwofunctionalparts:anactionpartandabranchdecisionpart,andpdo alsohasbothastartnodeandanendnode.hestatetransitionofpdoisbasedon stack and indexed memory. Since PDO has been successfully applied to image and sound classification problems, it can be said that PDO has a splendid ability for static problems. GNP is designed mainly to deal with problems in dynamic environments. First,themainconceptofGNPistomakeuseoftheimplicitmemoryfunction.herefore, GNP does not presuppose that it uses explicit memories such as stack and indexed memories. Second, GNP has judgment nodes and processing nodes which correspond to branch decision parts and action parts in PDO, respectively. Note that GNP separates judgment and processing functions, while both functions of PDO are in a node. herefore GNP can create more complex combinations/rules of judgment and processing.finally,thenodesofgnphaveauniquenodenumberandthenumberofnodes is the same in all the individuals. his characteristic contributes to executing RL effectively in GNP-RL(see section 3.6). Some information on the techniques of graph or networkbasedgpisgiveninlukeandspector(1996). Finally, we explain the methods that combine evolution and learning. In Downing (21), the special terminal nodes for learning are introduced to GP and the contents oftheterminalnodes,i.e.,theactionsofagentsaredeterminedbyqlearning.iniba (1998),aQtableisproducedbyGPtomakeanefficientstatespaceforQlearning,e.g., if the GP program is ( B( xy)(+z5)), it represents a 2-dimensional Q table having 2 axes, x yand z + 5.InKamioandIba(25),theterminalnodesofGPselectappropriateQtablesandtheagentactionisdeterminedbytheselectedQtable. hemost important difference between these methods and GNP-RL is how to create state-action spaces(q tables). GNP creates Q tables using its graph structures. Concretely speaking, an activated node corresponds to the current state, and selection of a function in the activated node corresponds to an action. In the other methods, the current state is determined by the combination of the inputs, and actions are actual actions such as moveforward,turnrightandsoon. 3 Genetic Network Programming In this section, Genetic Network Programming is explained in detail. GNP is an extension of GP in terms of gene structures. he original motivation for developing GNP is based on the more general representation ability of graphs as opposed to that of trees in dynamic environments. Evolutionary Computation Volume 15, Number 3 371

4 S.Mabu,K.HirasawaandJ.Hu gent gent fitness value Environment LIBRRY J1: P1: J2: P2: Jk:... Pl:... start node Directed graph structure P1 J1 B d1 1 2 d1 3 4 Node gene node i Ki IDi di Ci di B J2 P2 Gene structure node 1 node node 3 node Connection gene B B Ci di node B J: Judgement function P: Processing function : Judgement node : Processing node : ime delay Figure 1: Basic structure of GNP. 3.1 Basic Structure of GNP Components Fig. 1showsabasicstructureofGNP.GNPprogramiscomposedofonestartnode, plural judgment nodes and plural processing nodes. In Fig. 1, there are one start node, two judgment nodes and two processing nodes, and they are connected to each other. Startnodehasnofunctionandnoconditionalbranch.heonlyroleofthestartnode is to determine the first node to be executed. Judgment nodes have conditional branch decision functions. Each judgment node returns a judgment result and determines the next node to be executed. Processing nodes work as action/processing functions. For example, processing nodes determine the agent s actions such as go forward, turn right and turn left. In contrast to judgment nodes, processing nodes have no conditional branch. By separating processing and judgment functions, GNP can handle various combinations of judgment and processing. hat is, how many judgments and which kinds of judgment should be used can be determined by evolution. Suppose there areeightkindsofjudgmentnodes(j 1,...,J 8 ),andfourkindsofprocessingnodes (P 1,..., P 4 ).hen,gnpcanmakeanodetransitionbyselectingnecessarynodes,e.g., J 1 -> J 5 -> J 3 -> P 1. Here,itsaysthatjudgmentnodes J 1, J 5 and J 3 areneededfor processingnode P 1.Byselectingthenecessarynodes,GNPprogramcanbequitecompact and evolved efficiently. In this paper, as described above, each processing node determines an agent s actionsuchas goforward, turnright andsoon.ndeachjudgmentnodedetermines thenextnodeafterjudging whatisintheforward?, whatisintheright? andso on. However, in other applications, they could, for example, be applied to other functions such as judge sensor values(judgment), determine wheel speed(processing) of Khepera robot(by K-eam Corp.), judge whether stocks rise or drop(judgment) and determine buy or sell strategy(processing) in stock markets. 372 Evolutionary Computation Volume 15, Number 3

5 Genetic Network Programming and Its Extension GNP evolves the graph structure with the predefined number of nodes, so it never causesbloat 2. Inaddition,GNPhasanabilitytousecertainjudgment/processing nodesrepeatedlytoachieveatask. herefore,evenifthenumberofnodesispredefined and small, GNP can perform well by making effective node connections based on re-usingnodes.saresult,wedonothavetoprepareanexcessivenumberofnodes. he compact structure of GNP is a quite important and distinguishing characteristic, because it contributes to saving memory consumption and calculation time MemoryFunction henodetransitionbeginsfromastartnode,buttherearenoterminalnodes.fterthe start node, the current node is transferred according to the node connections and judgmentresults,inotherwords,theselectionofthecurrentnodeisinfluencedbythenode transitions of the past. herefore, the graph structure itself has an implicit memory function of the past agent actions. lthough a judgment node is a conditional branch decision function, the GNP program is not merely the aggregate of if-then rules, because it includes information of past judgment and processing. For example, in Fig. 1,afternode1(processingnode P 1 )isexecuted,thenextnodebecomesnode2(judgmentnode J 2 ).herefore,whenthecurrentnodeisnode2,wecanknowtheprevious processingwas P 1. henodetransitionofgnpendswhentheendconditionissatisfied,e.g.,when the time step reaches the preassigned one or the GNP program completes the given task imedelays GNPhastwokindsoftimedelays:thetimedelayGNPspendsonjudgmentorprocessing,andtheoneitspendsonnodetransitions. Inrealworldproblems,whenagents judge environments, prepare for actions and take actions, they need time. For example, whenamaniswalkingandseesapuddlebeforehim,hewillavoidit.tthatmoment, ittakessometimetojudgethepuddle(timedelayofjudgment),toputjudgmentinto action(time delay of transition from judgment to processing) and to avoid the puddle(timedelayofprocessing).sincetimedelaysarelistedineachnodegeneandare unique attributes of each node, GNP can evolve flexible programs considering time delays. Inthispaper,timedelayofeachnodetransitionissetatzerotimeunit,thatof eachjudgmentnodeisonetimeunit,thatofeachprocessingnodeisfivetimeunits, andthatofastartnodeiszerotimeunit.inaddition,theonestepofanagent sbehaviorisdefinedinsuchawaythatonestependswhenanagentusesfiveormoretime units.husanagentshoulddofewerthanfivejudgmentsandoneprocessing,orfive judgmentsinonestep.supposetherearethreeagents(agent,agent1,agent2)inan environment. During one step, first agent takes an action, next agent 1, finally agent 2. In this way, agents repeatedly take actions until reaching the maximum preassigned steps. nother important role of time delays and steps is to prevent the program from falling into deadlocks. For example, if an agent cannot execute processing because of thejudgmentloop,thenonestependsafterfivejudgments.suchaprogramisremoved from the population in the evolutionary process, or the node transition is changed by the learning process of GNP-RL, as described later. 2 phenomenonthataprogramsize,i.e.,thenumberofnodes,becomestoolargeasgenerationgoeson. Evolutionary Computation Volume 15, Number 3 373

6 S.Mabu,K.HirasawaandJ.Hu 3.2 GeneStructure he graph structure of GNP is determined by the combination of the following node genes.geneticcodeofnode i( i n 3 1)isalsoshowninFig.1. K i representsthenodetype, K i = meansstartnode, K i = 1meansjudgment nodeand K i = 2meansprocessingnode. ID i representstheidentificationnumber ofthenodefunction,e.g., K i = 1and ID i = 2meanthenodeis J 2. d i isthetime delayspentonjudgmentorprocessing. Ci, CB i...showthenodenumberconnected fromnode i. d i, db i,...meantimedelaysspentonthetransitionfromnode itonode Ci, CB i...,respectively.judgmentnodesdeterminetheuppersuffixoftheconnection genes to refer to depending on their judgment results. For example, if the judgment resultis B,GNPrefersto Ci B and d B i.however,astartnodeandprocessingnodes useonly Ci and d i,becausetheyhavenoconditionalbranch. 3.3 Initialization of a GNP Population Fig. 2 shows the whole flowchart of GNP. n initial population is produced according tothefollowingrules. First,wedeterminethenumberofeachkindofnode 4,thereforeallprogramsinapopulationhavethesamenumberofnodesandthenodeswith the same node number have the same function. However, the extended algorithm, GNP-RL described later, determines the node functions automatically, so we only need to determine the number of judgment nodes and processing nodes, e.g., 4 judgment nodesand2processingnodes.heconnectiongenes C i, CB i,...aresetatthevalues selectedrandomlyfrom 1,..., n 1(except iinordertoavoidself-loop). 3.4 RunofaGNPProgram henodetransitionofgnpisbasedon C i.ifthecurrentnode iisajudgmentnode, GNPexecutesthejudgmentfunction ID i anddeterminesthenextnodeusingitsresult. Forexample,whenthejudgmentresultis B,thenextnodebecomes C B i.whenthe currentnodeisaprocessingnode,afterexecutingtheprocessingfunction ID i,thenext nodebecomes C i. 3.5 GeneticOperators In each generation, the elite individuals are preserved and the rest of the individuals are replaced with the new ones generated by crossover and mutation. In Simulation I(Section 4.3), first, 179 individuals are selected from the population by tournament selection 5 andtheirgenesarechangedbymutation. hen,12individualsarealso selected from the population and their genes are exchanged by crossover. Finally, the 299 individuals generated by mutation and crossover and one elite individual form the next population Mutation Mutationisexecutedinoneindividualandanewoneisgenerated[Fig.3].heprocedure of mutation is as follows. 3 Eachnodeinaprogramhasauniquenodenumberfrom to n 1(n:totalnumberofnodes),respectively. 4 Fiveofeachkindinthispaper.Itcouldbedeterminedexperimentally,howeverinthispaper,previous experienceindicatesfivenodespereachkind(j 1, J 2,..., P 1, P 2,...)couldkeepareasonablebalanceof expression ability and search speed. 5 hecalculationcostoftournamentselectionisrelativelysmall,becauseitsimplycomparesthefitness values of some individuals, and we can easily determine selection pressure by tournament size. hus, we use tournament selection in this paper. ournament size is set at six. 374 Evolutionary Computation Volume 15, Number 3

7 Genetic Network Programming and Its Extension start generate an initial population ind=1 Judgement/Processing No ind=ind+1 No trial ends? Yes ind=the number of individuals? Yes reproduction mutation crossover No ind : individual number last generation? Yes stop Figure 2: Flowchart of GNP system. 1. Select one individual using tournament selection and reproduce it as a parent. 2.Eachconnectionofeachnode(C i )isselectedwiththeprobabilityof P m.heselected C i ischangedtoothervalue(nodenumber)randomly. 3. Generated new individual becomes the new one of the next generation Crossover Crossover is executed between two parents and generates two offspring[fig. 4]. he procedure of crossover is as follows. 1. Select two individuals using tournament selection twice and reproduce them as parents. 2.Eachnode iisselectedasacrossovernodewiththeprobabilityof P c. 3. wo parents exchange the genes of the corresponding crossover nodes, i.e., the nodes with the same node number. 4. Generated new individuals become the new ones of the next generation. Fig. 4 shows a crossover example of the graph structure with three processing nodes for simplicity. If GNP exchanges the genes of judgment nodes, it must exchange allthegeneswithsuffix, B, C,...simultaneously. 3.6 Extended lgorithm GNP with Reinforcement Learning(GNP-RL) In this subsection, we propose an extended algorithm, GNP with Reinforcement Learning(GNP-RL). Standard GNP(SGNP) described in the previous section is based on the Evolutionary Computation Volume 15, Number 3 375

8 S.Mabu,K.HirasawaandJ.Hu 1 C3=1 mutation Each branch is selected with the probability of Pm 2 3 C3=2 he selected branch becomes connected to another node randomly. Figure 3: Mutation. general evolutionary framework such as selection, crossover and mutation. GNP-RL is based on evolution and reinforcement learning(sutton and Barto, 1998). he aim of combining RL with evolution is to produce programs using the current information (state and reward) during task execution. Evolution-based methods change their programs mainly after task execution or enough trials, i.e., offline learning. On the other hand, GNP-RL can change its programs incrementally based on rewards obtained during task execution, i.e., online learning. For example, when an agent takes a good action withapositiverewardatacertainstate,theactionisreinforcedandtheactionwillbe adopted with higher probability when visiting the state again. Online learning is one of the advantages of GNP-RL. he other advantage is a combination of a diversified search of evolution and an intensified search of RL. he role of evolution is to make rough structures, i.e., plural paths of node transition, through selection, crossover and mutation. he role of RL is to determine one appropriate path in a structure made by evolution. Because RL is executed based on immediate rewards obtained after taking actions, intensified search, i.e., local search, can be executed efficiently. Evolution changes programs largely than RL and the programs(solutions) could escape from local minima, so we call evolution as a diversified search. henodesofgnphaveauniquenodenumberandthenumberofnodes(states)is the same in all the individuals. In addition, the crossover operator exchanges the nodes withthesamenodenumber.herefore,thelargechangesoftheqtablesdonotoccur and the obtained knowledge in the previous generation can be used effectively in the current generation Basic Structure of GNP-RL Fig. 5 shows a basic structure of GNP-RL. he difference between GNP-RL and SGNP iswhetherornotpluralfunctionsexistinanode.eachnodeofsgnphasonefunction, butthatofgnp-rlhasseveralfunctionsandoneofthemisselectedbasedonapolicy. K i representsthenodetype,whichisthesameasinsgnp. ID ip (1 p m i 6 )showstheidentificationnumberofthenodefunction.infig.5, m i ofallnodesare setat2,i.e.,gnpcanselectthenodefunction ID i1 or ID i2. Q ip isaqvaluewhichis assigned to each state and action pair. In reinforcement learning, state and action 6 m i (1 m i M M:Maximumnumberoffunctionsinanode,e.g., M = 4)showsthenumberof nodefunctionsgnpcanselectatthecurrentnode i. m i isdeterminedrandomlyatthebeginningofthefirst generation, but they can be changed by mutation. 376 Evolutionary Computation Volume 15, Number 3

9 Genetic Network Programming and Its Extension parent1 1 1 parent2 d3 C3=1 * d3 2 3 ID3 2 C3=2 * 3 ID3 d3 d3* Each node is selected with the probability of Pc (crossover node) crossover offspring1 1 1 offspring2 d3 C3=1 * d C3=2 * * ID3 ID3 * d3 d3 Figure 4: Crossover. must be defined, and generally, the current state is determined by the combination of the current information, e.g., sensor inputs, and action is an actual action an agent takes, e.g., go forward. However, in GNP-RL, the current state is defined as the current node, andaselectionofanodefunction(id ip )isdefinedasanaction. d ip isthetimedelay spentonjudgmentorprocessing. C ip, CB ip,...showthenodenumberofthenextnode. d ip, db ip,...meantimedelaysspentonthetransitionfromnode itonode C ip, CB ip,..., respectively Run of GNP with Reinforcement Learning he node transition of GNP-RL also starts from a start node and continues depending on the node connections and judgment results. Ifthecurrentnode iisajudgmentnode, first, oneqvalueisselectedfrom Q i1,..., Q imi based on ε-greedy policy. hat is, a maximum Q value among Q i1,..., Q imi isselectedwiththeprobabilityof 1 ε,orarandomoneisselectedwith theprobabilityof ε,thenthecorresponding ID ip isselected.gnpexecutestheselected judgmentfunction ID ip anddeterminesthenextnodedependingonthejudgmentresult.forexample,iftheselectedfunctionis ID i2 andthejudgmentresultis B,the nextnodebecomesnode Ci2 B. If the current node is a processing node, GNP selects and executes a processing functioninthesamewayasjudgmentnodes,andthenextnodebecomesnode Ci2 whentheselectedfunctionis ID i2. Here, a concrete example of node transition is explained using Fig. 6. he first nodeisajudgmentnode2andtherearefunctionsjfandd(seeable1).supposejf is selected based on the ε-greedy policy and the judgment result is D(=floor). hen the nextnodenumberbecomes C21 D = 4.Innode4,theprocessingfunctionMFisselected, Evolutionary Computation Volume 15, Number 3 377

10 S.Mabu,K.HirasawaandJ.Hu Directed graph structure start node node Node gene Gene structure 1 node Connection gene 4 node node 3 node node i Ki IDi1 Qi1 di1 Ci1 Node gene... Connection gene di1 di1... IDimi Qimi dimi... Ci1 B B Cimi dimi B Cimi dimi B Judgement node in the case of mi=2 : ime delay Qi1 di1 IDi1 IDi2 Qi2 di2 di1 node i node Ci1 B di1 di2... node Ci1 One branch is selected according to the judgement result. B di2 node node B Ci2 B Ci2 Processing node node i Qi1 di1 IDi1 di1... node j node Ci1 IDi2 Qi2 di2 di2 node Ci2 Figure 5: Basic structure of GNP with Reinforcement Learning. soanagentmovesforward,andthenextnodebecomesnode C 42 = GeneticOperators crossoveroperatoringnp-rlisthesameasinsgnp,i.e.,allthegenesoftheselected nodes are exchanged. However, GNP-RL has its own mutation operators. he procedure is as follows. 1. Select one individual using tournament selection and reproduce it as a parent 2. Mutation operator here are three kinds of mutation operators[fig. 7], and, uniformly selected, one operator is executed. (a) connection of functions: Each node connection is re-connected to another node(c ip ischangedtoanothernodenumber)withtheprobabilityof P m. (b)contentoffunctions:eachfunctionisselectedwiththeprobabilityof P m and changedtoanotherfunction,i.e., ID ip and d ip areeachchanged. 378 Evolutionary Computation Volume 15, Number 3

11 Genetic Network Programming and Its Extension node 2 ID21=1 Q21=1. JF D ID22=5 Q22=.1... tile hole obstacle agent floor C21=1 B C21=5 C C21=8 E C21=1 D C21=4 node 4 ID41=1 Q41=.2 L MF ID42=3 Q42=.5 C41=7 C42=9 JF : judge forward D : direction of the nearest ile from the agent MF: move forward L : turn left Figure 6: n example of node transition using the nodes for ileworld problem. (c)numberoffunctions:eachnode iisselectedwiththeprobabilityof P m,and thenumberoffunctions m i ischangedto 1,...,or Mrandomly.Iftherevised m i becomeslargerthantheprevious m i,thenoneormorenewfunctionsselectedfrom LIBRRYareaddedtothenodesothatthenumberoffunctions becomestherevised m i.iftherevised m i becomessmaller,thenoneormore functions are deleted from the node. 3. he generated new individual becomes the new one of the next generation. 3.7 LearningPhase Reinforcement learning is carried out when agents are carrying out their tasks and terminates when the time step reaches the predefined steps. he learning phase of GNPisbasedonabasicSarsaalgorithm(SuttonandBarto,1998).SarsacalculatesQ valueswhicharefunctionsofstate sandaction a. Qvaluesestimatethesumofthe discountedrewardsobtainedinthefuture.supposethatanagentselectsanaction a t atstate s t attime t,areward r t isobtainedandanaction a t+1 istakenatthenextstate s t+1.hen Q(s t, a t )isupdatedasfollow. Q(s t, a t ) Q(s t, a t ) + α [r t + γq(s t+1, a t+1 ) Q(s t, a t )] (1) αisastepsizeparameter,and γisadiscountratewhichdeterminesthepresent valueoffuturerewards:arewardreceived ktimestepslaterisworthonly γ k 1 times oftherewardsupposedtoreceiveatthecurrentstep. sdescribedbefore,astatemeansthecurrentnodeandanactionmeanstheselectionofafunction.hereaprocedureforupdatingqvalueisexplainedusingfig.8 which shows states, actions and an example of node transition. 1.ttime t,gnprefersto Q i1, Q i2,...,q imi andselectoneofthembasedon ε- greedy.supposethatgnpselects Q ip andthecorrespondingfunction ID ip. 2.GNPexecutesthefunction ID ip,getsthereward r t andthenextnode jbecomes C ip. 3.ttime t + 1,GNPselectsoneQvalueinthesamewayasstep1.Supposethat Q jp isselected. Evolutionary Computation Volume 15, Number 3 379

12 S.Mabu,K.HirasawaandJ.Hu 1 C31=1 2 C32=2 3 mutation (the connection) 1 C31=2 2 C32=2 3 he connection is changed randomly. IDi di node i mutation (the content of functions) * * IDi di he content of the functions is changed randomly. node i mi=2 mutation (the number of functions) mi=3 he number of functions is selected from 1,..., M. Figure 7: Mutation of GNP with Reinforcement Learning. 4.Qvalueisupdatedasfollows. Q ip Q ip + α [r t + γq jp Q ip ] 5. t t + 1, i j, p p thenreturnstep2. Inthisexample,node iisaprocessingnode,butifitisajudgmentnode,thenext currentnodeisselectedamong C ip, CB ip,...dependingonthejudgmentresult. 4 Simulations o confirm the effectiveness of the proposed method, the simulations for determining the agents behavior using the ileworld problem(pollack and Ringuette, 199) are described in this section. 4.1 ileworld ileworldiswellknownasatestbedfortheproblemofagents.fig.9showsanexample of ileworld, which is a 2D grid world including multi-agents, obstacles, tiles, holes and floors.gentscanmovetoacontiguouscellinonestep.moreover,agentscanpusha 38 Evolutionary Computation Volume 15, Number 3

13 Genetic Network Programming and Its Extension node i reward rt node j (=Cip ) reward rt+1 Qi1 Qj1 IDi1 IDj1 action at Qip IDip Qimi Cip Qjp at+1 IDjp Qjmj IDimi IDjmj state st state st+1 t t+1 time Figure 8: n example of node transition. tiletothenextcellexceptwhenanobstacleorotheragentexistsinthecell.whenatile isdroppedintoahole,theholeandthetilevanish,i.e.,theholeisfilledwiththetile. gentshavesomesensorsandactionabilities,andtheiraimistodropmanytilesinto holes as fast as possible. herefore, agents are required to use sensors and take actions properly according to their situations. Since the given sensors and simple actions are not enough to achieve tasks, agents must make clever combinations of judgment and processing. henodesusedbyagentsareshowninable1.hejudgmentnodes {JF,JB,JL, JR }return {tile,hole,obstacle,floor,agent },and {D,HD,HD,SD }return { forward,backward,left,right,nothing }asjudgmentresults,like, B,...inFig. 1. Fig.1showsthefourdirectionsanagentcanperceivewhenitfacesnorth Fitness and Reward trialendswhenthetimestepreachesthepreassignedstep,andthenfitnessiscalculated. Fitness is used in the evolutional processes and Reward is used in the learning phase of GNP-RL. Fitness = the number of dropped tiles Reward=1(whenanagentdropsatileintoahole) 4.2 SimulationConditions he simulation conditions are shown in able 2. For the comparison, the simulations arecarriedoutbysgnp,gnp-rl,standardgp,gpwithdfs,andepevolvingfsms Conditions of GNP sshowninable2,thenumberofnodesinaprogramis61.inthecaseofsgnp,the numberofeachkindofjudgmentandprocessingnodesisfixedatfive.inthecaseof GNP-RL,4judgmentnodesand2processingnodesareused,butthenumberofdifferentkindsofnodes(ID i )arechangedthroughtheevolution.tthefirstgeneration, allnodeshavefunctionsrandomlyselectedfromlibrry.but, ID i areexchanged by crossover and also changed by mutation, thus the appropriate kinds of nodes are selected as a result of evolution. Evolutionary Computation Volume 15, Number 3 381

14 S.Mabu,K.HirasawaandJ.Hu (North) Forward ile Hole Obstacle Floor gent Left Left Forward Right Figure 9: ileworld. Backward Backward (South) Figure 1: Four directions gents perceive. he number of elite individuals and offspring generated by crossover and mutationispredefined.simulationiusestheenvironmentshowninfig.9,wherethepositions of tiles and holes are fixed, so the same environment is used every generation. Ontheotherhand,inSimulationII,thepositionsoftilesandholesaredetermined randomly, so the problem becomes more difficult and complex. In Simulation I, the best individual is preserved as an elite one, but in Simulation II, five elite individuals are preserved, because the environment changes generation by generation. In fact, thebestindividualinthepreviousgenerationdoesnotalwaysshowagoodresultin the current generation. herefore, in order to make the performance of each method stable, we preserve five good individuals in Simulation II. In GNP-RL, when creating offspring in the evolution, offspring inherits the Q values of its parents and uses them asinitialvalues.hatis,theoffspringgeneratedbymutationhasthesameqvaluesas theparentsbecausemutationdoesnotoperatesonqvalues.hence,theqvaluesofan elite individual carry over to the next generation. Furthermore, the offspring generated bycrossoverhastheexchangedqvaluesoftheparentsinaddition,threeagentsinthe tileworld share the Q values. Crossoverrate P c andmutationrate P m aredeterminedappropriatelythrough our experiments, which maintains the variation of the population, but does not change theprogramstoomuch.hesettingsoftheparametersusedinthelearningphaseof GNP-RLareasfollows.hestepsizeparameter αissetat.9inordertofindsolutions quicklyandthediscountrate γissetat.9inordertosufficientlyconsiderthefuture rewards. ε is set at.1 experimentally, which considers the balance between exploitation and exploration. In fact, the programs with lower epsilon fall into local minima with higher probability, and those with higher epsilon take too much random actions. M(themaximumnumberoffunctionsinanode)issetatthebestvalueamong M = 2, 3and Conditions of GP WeuseGPasadecisionmaker,sothefunctionnodesareusedas if-thentypebranch decision functions, and the terminal nodes are used as action selection functions. he terminal nodes of standard GP are composed of the processing nodes of GNP: {MF, 382 Evolutionary Computation Volume 15, Number 3

15 Genetic Network Programming and Its Extension able 1: Function Set. Judgment node J symbol content J 1 JF judgeforwrd J 2 JB judgebckwrd J 3 JL judgelefside J 4 JR judgerighside J 5 D directionofthenearestilefromtheagent J 6 HD directionofthenearestholefromtheagent J 7 HD directionofthenearestholefromthenearestile SD directionofthesecondnearestilefromtheagent J 8 Processing node P symbol content P 1 MF moveforward P 2 R turnright P 3 L turnleft P 4 S stay L,R,S},andthefunctionnodesarethejudgmentnodesofGNP: {JF,JB,JL,JR, D, HD, HD, SD}. erminal nodes have no arguments and function nodes have five arguments corresponding to the judgment results. In the case of GP with DFs, the maintreeuses {DF1,...,DF1 7 }asterminalnodesinadditiontotheterminal andfunctionnodesinstandardgp.hedftreeusesthesamenodesasstandard GP.hegeneticoperatorsofGPusedinthispaperarecrossover(PoliandLangdon, 1998, 1997), mutation and inversion(koza, 1992). In the simulations, the maximum depthofthetreesisfixedinordertoavoidbloat,butthesettingofthemaximumdepth is very important, because the expression ability is improved as the depth becomes large, while the search space is increased. herefore, we try to use various depths in therangepermittedbymachinememoryandcalculationtime,andusea full anda ramped half-and-half initialization methods(koza, 1992) in order to produce trees with various sizes and shapes in the initial population Conditions of EP EPusesthesamesensorinformationasthejudgmentnodesofGNPuse,andtheoutputsarethesameasthecontentsoftheprocessingnodes. GenerallyEPmustdefine transitions and outputs for all combinations of states and inputs. Here, we would like todiscusshowthecomplexityoftheepandgnpprogramsdiffersdependingonthe problem. able 3 shows the number of outputs/connections for each individual in EP andgnp.fig. 11showsthenumberofoutputsateachstate/node. Incase1,there is only one sensor which can distinguish two objects, and the number of states/nodes is6 8. hen,thenumberofoutputsofepbecomes12,thatofsgnpbecomes1, and that of GNP-RL becomes 1-4(variable). However, as the number of sensors 7 henumberofdfsineachindividualis1andeachdfiscalledbytheterminalnodesofthemain tree. 8 startnodeofgnpisnotcountedbecauseithasonlyonebranchdeterminingthefirstjudgmentor processingnodeanddoesnothaveanyfunctions. EPhasabranchdeterminingafirststate,butitisnot counted as an output. Evolutionary Computation Volume 15, Number 3 383

16 S.Mabu,K.HirasawaandJ.Hu able 2: Simulation conditions. GNP-RL SGNP GP EP GP-DFs Population size crossover mutation 179[175] 179[175] 119[115] 299[295] inversion 6 elite 1[5] 1[5] 1[5] 1[5] program size GP: 3 5 state 6,3,5 GP-DFs:main3,4,DF2,3 input1 4 Crossoverrate P c.1.1 Mutationrate P m.1.2[.1].1[.1].1 ournament size 6 Otherparameters α =.9 γ =.9 ε =.1 M = 4 [3] []: conditions in Simulation II Population size: the number of individuals Program size: the number of nodes including one start node(gnp), max depth(gp), max number of states and number of inputs(ep). processing node one connection Y connections state X Y outputs judgement node GNP EP Figure 11: he number of outputs from a node/state. (inputs) and the number of objects each sensor can distinguish increase, the number of outputs of EP becomes exponentially large(case 2 case 4). On the other hand, the number of connections in GNP does not become exponentially large because each judgment nodedealswithonlyonesensoranddoesnotneedtoconsiderallthecombinationsof the inputs. Case4inable3showsthecaseoftheileworldproblem: thetotalnumberof outputsofanepprogramis 23, 437, 5(= ).hisisimpracticaltouse,sowe limitthenumberofsensors(inputs)usedateachstatetoacertainnumber(seeable2). However,whichinputsareusedateachstateisaveryimportantmatter,thus,inorder to find the necessary kind(s) of input(s), the additional mutation operator is introduced, whichcanchangethekind(s)ofinput(s)usedateachstate.fig.12showsanexample oftheepprogramusingtwostatesandonesensorateachstateforsimplicity.inthis example,theinputusedatstate1isthejudgmentresultofjfandthatofstate2is thejudgmentresultofd.hen,thenumberoftransitionsandoutputsisfive,each corresponding to the judgment result, and each output shows the next action an agent 384 Evolutionary Computation Volume 15, Number 3

17 Genetic Network Programming and Its Extension able3:herelationbetweenthenumberofinputsandoutputsinepandgnp. case1 case2 case3 case4 X Y EP ,36 23,437,5 Z SGNP GNP-RL(M = 4) 1-4(variable) X: the number of inputs(sensors) Y:thenumberofobjectseachsensorcandistinguish Z: the total number of outputs(connections) the number of states/nodes: 6(judgment node: 4, processing node: 2 in the case of GNP) initial state tile/mf agent/r hole/r right/r floor/mf 1 nothing/mf 2 left/l obstacle/l JF D forward/mf backward/r in the case where one input is dealt with Figure12:nexampleofanEPprogram. shouldtake.ifthenumberofinputsateachstateistwoinsteadofone,thenumberof transitions and outputs becomes 25 each. t the beginning of the first generation, the predefined number of sensors are assignedtoeachstaterandomly,butthetypesofsensorsarechangedbythemutation operatorasthegenerationgoeson. sshowninable2,thenumberofsensors(inputs)andthemaximumnumberofstatesaresetatsomevalueinordertofindgood settings for EP. he mutation operators used in the simulations are {DD SE, REMOVE SE, CHNGE RNSIION, CHNGE OUPU, CHNGE INIIL SE, CHNGE INPU}. he last operator is the additional one adopted in this paperandtheothersarethesameastheonesusedinngeline(1994). 4.3 SimulationI SimulationIusestheenvironmentshowninFig. 9wherethereare3tilesand3 holes.hreeagentshavethesameprogrammadebygnp,gporep.inthisenvironment, since each input for judgment nodes is not the complete information needed to distinguish various situations, each method is required to judge its situations and take actions properly by combining various kinds of judgment and processing nodes. he maximumnumberoftimestepsissetat15. Evolutionary Computation Volume 15, Number 3 385

18 S.Mabu,K.HirasawaandJ.Hu able4:hedataonthefitnessofthebestindividualsatthelastgenerationinsimulation I. GNP-RL SGNP GP-DFs GP EP average standard deviation t-test GNP-RL (p-value) SGNP Figs. 13,14and15showthefitnesscurvesofthebestindividualsateachgeneration averaged over 3 independent simulations. From the results, GNP-RL shows the best fitness value at 5 generation. In early generations, SGNP exhibited better fitness,becauseqvaluesingnp-rlaresetatzerointhefirstgenerationandmust be updated gradually. However, GNP-RL can produce the better result in the later generations by appropriately learned Q values. able 4 shows the averaged fitness values of the best individuals over 3 simulations 9 atthelastgeneration,theirstandarddeviation,andtheresultsofat-test(onesided test). he results of the t-test show the p-values between GNP-RL and the other methods, and between SGNP and the other methods. here are significant differences betweengnp-rlandtheothermethods,andbetweensgnpandgp,gpwithdfs, and EP. lthough it seems natural that the method using RL can obtain better solutions thantheothermethodswithoutit,theaimofdevelopinggnp-rlistosolveproblems fasterthantheotherswithinthesametimelimitofactions(samesteps).inotherwords, GNP-RL aims to make full use of the information obtained during task execution in order to make the appropriate node transitions. FromFig.14,weseethatstandardGPofdepthfour,initializedbythefullmethod (GP-full4), shows better results than the other standard GP programs, and GP with DFsofdepththree(maintree)anddepthtwo(DFtree)initializedbythefullmethod (GP-DF-full3-2) produces the best result of all the GP programs. However, in this problem,thearityofthefunctionnodesofgpisrelativelylarge(five),sothetotal numberofnodesofgpbecomesquitelargeasthedepthbecomeslarge. Forexample, GP-full4 has 781 nodes, GP-full5 has 3,96 nodes, and GP-full6 has 19,531 nodes. lthough GP programs can have higher expression ability as the number of nodes increases, they take much time to explore the programs and much memory is needed. For example, GP-full6 takes too much time to execute the programs and GP(depth7) cannot be executed because of the lack of memory in our machine(pentium4 2.5GHz, DDR-SDRM PC21 512MB). On the other hand, GNP can obtain good results using relatively small number of nodes. sshowninfig.15,epusingthreeinputsandfivestatesshowsbetterresults,so thissettingissuitablefortheenvironment.epusesagraphstructure,soitcanalsoexecute state transition considering the past agents actions. Furthermore, as the number of states increases, EP can implicitly memorize the past action sequences. However, if therearemanyinputs,itcausesalargenumberofoutputsandstatetransitionrulesand the programs then become impractical for exploration and execution. he structure of GNP does not become exponentially large even if the number of inputs increases as 9 theresultsofthebestsettings,i.e.,gp-full4,gp-df-full3-2andep-input3-state5.able5,6and7also show the results of the best settings. 386 Evolutionary Computation Volume 15, Number 3

19 fitness fitness fitness Genetic Network Programming and Its Extension 22 2 GNP-RL (21.23) SGNP (18.) fitness at the last generation generation Figure 13: Fitness curves of GNP in Simulation I GP-DF-full3-2 (15.43) GP-DF-ramp4-3 (13.5) GP-full4 (14.) GP-DF-full4-3 (14.46) GP-DF-ramp3-2 (13.86) GP-full5 (13.76) GP-ramp4 (1.1) GP-ramp5 (1.3) generation Figure14:FitnesscurvesofGPinSimulationI EP-input4 state 5 (14.93) EP-input1 state 6 (13.3) EP-input3 state 5 (16.3) EP-input2 state 3 (13.7) generation Figure15:FitnesscurvesofEPinSimulationI. Evolutionary Computation Volume 15, Number 3 387

20 S.Mabu,K.HirasawaandJ.Hu able 5: Calculation time for 5 generations in Simulation I. GNP-RL SGNP GP-DFs GP EP Calculation time[s] 1,364 1,19 3,252 3, ratio of each method to SGNP the average number of functions in each node generation Figure16:Changeoftheaveragenumberoffunctions m i ineachnodeingnp-rl. describedin4.2.3,thereforemanymorestates(nodes)canbeusedingnpthaninep. s a result, the implicit memory function of GNP becomes more effective in dynamic environments than that of EP. able 5 shows the calculation time for 5, generations. SGNP is the fastest, GNP- RLissecondandEPisthird.GNP-RLtakesmoretimethanSGNP,becauseitexecutes RLduringtasks,however,itdoesnottakesomuchtime. hemaximumnumberof functions(m)ineachnodeis4andoneofthemisselectedasanaction;thisprocedure doesnottakemuchtime. Inaddition,moreimportantly, m i tendstodecreaseasthe generationgoesonasshowninfig.16(1.75atthelastgeneration)becausetheappropriate number and contents of functions are selected automatically in the evolutional processes. herefore, reinforcement learning just selects one function from 1.75 functions on average. his tendency contributes to saving time. In addition, the relatively large epsilon (=.1) succeeded in achieving the tasks thanks to this tendency, because lessthantwofunctions(actions)areinanodeonaverageatthelastgeneration,while the relatively large epsilon is useful at the beginning of the generations, because the agentscantrytodomanykindsofactionsandfindgoodonesbyrl.eptakesmore timethangnpandgnp-rl,butitcansaveitscalculationtimecomparedwithordinaryep,becausethenumberofinputsislimitedtothreeandthestructurebecomes compact. ctually,thecalculationtimeofepusingfourinputsis4,184. GPandGP withdfshavemanynodes,thustheytakemoretimethantheothersinevolutional processes. Fig.17(a)showsthetypicalnodetransitionoftheupperleftagent(inFig.9)operated by SGNP. he x-axis shows the symbols of the nodes, and the y-axis distinguishes thesamekindofnodes,i.e.,therearefivenodespereachkind/symbol 1 andtheyhave thenumber,1,2,3and4.forexample, (x, y) = (4, 1)showsthesecondJFnode.From thefigure,wecanseethatthespecificnodesarerepeatedlyused. 1 FiveJFnodes,fiveJBnodes,...,fiveMFnode,fiveLnodes,...areusedinSGNP. 388 Evolutionary Computation Volume 15, Number 3

21 Genetic Network Programming and Its Extension MF L R S JF JB JL JR D HD HDSD (a)wholenodetransitionfor15steps reward (drop tile) judgement result floor forward MF JF D MF JF (x,y)=(,1) (4,1) (8,) (,1) (4,1) obstacle reward right forward backward D MF R HD MF HD (8,) (,) (2,) (9,2) (,2) (1,4) (b) partial node transition extracted from(a) Figure 17: Node transition of standard GNP. Fig. 17(b) shows the partial node transition extracted from the whole node transition[fig.17(a)].infig.17(b),thefirstnodeismf(,1),sotheagentmovesforward, andthenextnodebecomesjf(4,1)accordingtothenodeconnectionfrommf(,1). husthepoints(,1)and(4,1)infig. 17(a)areconnectedwithline. Next,thejudgment JF(4,1) is executed and the judgment result is floor, thus the corresponding node branch(connectedtod(8,))isselected.henthepoints(4,1)and(8,)infig.17(a) are connected. fter executing the judgment D(8,)(judgment result: forward), the agent goes forward, judges forward(judgment result: obstacle), judge the tile direction (judgment result: right), and so on. Finally,thesimulationsusingEnvironment,BandC[Figs. 18,19and2]are carriedout.heconditionofeachmethodisthesameasthatshowingthebestresultin thepreviousenvironment[fig.9].fromfigs.21,22and23,gnp-rlandsgnpshow better results than the other methods. Evolutionary Computation Volume 15, Number 3 389

22 S.Mabu,K.HirasawaandJ.Hu Figure 18: Environment. Figure 19: Environment B. Figure 2: Environment C. 4.4 SimulationII In Simulation II, we use an environment whose size(2x2) and distribution of the obstaclesarethesameassimulationi.however,2tilesand2holesaresetatrandom positions at the beginning of each generation. In addition, when an agent drops a tile intoahole,thetileandtheholedisappear;however,anewtileandanewholeappear at random positions. herefore, the individuals obtained in the previous generation are required to show good performance in the new, unexperienced environment. his problem is more dynamic and suitable than Simulation I in terms of confirming the generalization ability of each method. he maximum number of steps is set at 3. Figs.24,25 11 and26showtheaveragedfitnesscurvesofthebestindividualsover 3 independent simulations at each generation. From the figures, we can see that GNP- RL can obtain the highest fitness value at the last generation because the information obtained during task execution is used for making node transitions efficiently. From able 6, we can also see that there are significant differences between GNP-RL and the other methods. SGNP can obtain better fitness value than GP, GP-DFs at the last generation, but, from able 6, it is found that there is no significant difference between SGNPandEP-input1-state6.InthecaseofEP,itisinterestingtonotethattheprograms insimulationiishowtheoppositeresultstothoseinsimulationi,i.e.,theprogramusing one input shows better results in Simulation II, while that using three inputs show better results in Simulation I. herefore, for EP in this environment, it is recommended thatanactionbedeterminedbyoneinputandthatarelativelylargenumberofstatesis used. In other words, this EP makes many simple rules and combines them considering the past state transition. his special structure of EP is similar to that of GNP. However, the advantage of GNP is automatically selecting the necessary number of inputs and actions depending on the situations, and moreover, it is found that GNP programs with 61nodesshowgoodresultsinbothSimulationIandII,thereforewearenotworried about the settings of the number of nodes. In fact, there are significant differences betweensgnpandep-input3-state5(p-value= )whichshowsthebestfitness valueinsimulationi,ep-input2-state3( )andep-input4-state5( ). InthecaseofGP,itisdifficulttofindeffectiveprograms,becausetheenvironment changes randomly generation by generation. In addition, GP has relatively complex structuresandwidesearchspacecomparedtognpandep,thusitismoredifficultfor GP to explore solutions. 11 Fig. 25showsthefitnesscurvesofGP-full5andGP-DF-full3-2,andthefitnessvaluesoftheother settings at the last generation. Because the fitness curves of the GP overlap each other, the best two results (GP-full5 and GP-DF-full3-2) are shown. 39 Evolutionary Computation Volume 15, Number 3

A Structural Optimization Method of Genetic Network Programming. for Enhancing Generalization Ability

A Structural Optimization Method of Genetic Network Programming. for Enhancing Generalization Ability International Journal of Engineering Innovation and Management Vol.6, No.2, 2016 A Structural Optimization Method of Genetic Network Programming for Enhancing Generalization Ability Shingo Mabu, Yamaguchi

More information

K-CUT CROSSOVER USING GRAPH THEORY IN GENETIC NETWORK PROGRAMMING. Received December 2011; revised April 2012

K-CUT CROSSOVER USING GRAPH THEORY IN GENETIC NETWORK PROGRAMMING. Received December 2011; revised April 2012 International Journal of Innovative Computing, Information and Control ICIC International c 2013 ISSN 1349-4198 Volume 9, Number 2, February 2013 pp. 641 650 K-CUT CROSSOVER USING GRAPH THEORY IN GENETIC

More information

Genetic Programming Prof. Thomas Bäck Nat Evur ol al ut ic o om nar put y Aling go rg it roup hms Genetic Programming 1

Genetic Programming Prof. Thomas Bäck Nat Evur ol al ut ic o om nar put y Aling go rg it roup hms Genetic Programming 1 Genetic Programming Prof. Thomas Bäck Natural Evolutionary Computing Algorithms Group Genetic Programming 1 Genetic programming The idea originated in the 1950s (e.g., Alan Turing) Popularized by J.R.

More information

1. Introduction. 2. Motivation and Problem Definition. Volume 8 Issue 2, February Susmita Mohapatra

1. Introduction. 2. Motivation and Problem Definition. Volume 8 Issue 2, February Susmita Mohapatra Pattern Recall Analysis of the Hopfield Neural Network with a Genetic Algorithm Susmita Mohapatra Department of Computer Science, Utkal University, India Abstract: This paper is focused on the implementation

More information

Genetic programming. Lecture Genetic Programming. LISP as a GP language. LISP structure. S-expressions

Genetic programming. Lecture Genetic Programming. LISP as a GP language. LISP structure. S-expressions Genetic programming Lecture Genetic Programming CIS 412 Artificial Intelligence Umass, Dartmouth One of the central problems in computer science is how to make computers solve problems without being explicitly

More information

Mining Equalized Association Rules from Multi Concept Layers of Ontology Using Genetic Network Programming

Mining Equalized Association Rules from Multi Concept Layers of Ontology Using Genetic Network Programming Mining Equalized Association Rules from Multi Concept Layers of Ontology Using Genetic Network Programming Guangfei Yang, Kaoru Shimada, Shingo Mabu, Kotaro Hirasawa and Jinglu Hu Abstract In this paper,

More information

Using Genetic Algorithms to optimize ACS-TSP

Using Genetic Algorithms to optimize ACS-TSP Using Genetic Algorithms to optimize ACS-TSP Marcin L. Pilat and Tony White School of Computer Science, Carleton University, 1125 Colonel By Drive, Ottawa, ON, K1S 5B6, Canada {mpilat,arpwhite}@scs.carleton.ca

More information

Suppose you have a problem You don t know how to solve it What can you do? Can you use a computer to somehow find a solution for you?

Suppose you have a problem You don t know how to solve it What can you do? Can you use a computer to somehow find a solution for you? Gurjit Randhawa Suppose you have a problem You don t know how to solve it What can you do? Can you use a computer to somehow find a solution for you? This would be nice! Can it be done? A blind generate

More information

Topological Machining Fixture Layout Synthesis Using Genetic Algorithms

Topological Machining Fixture Layout Synthesis Using Genetic Algorithms Topological Machining Fixture Layout Synthesis Using Genetic Algorithms Necmettin Kaya Uludag University, Mechanical Eng. Department, Bursa, Turkey Ferruh Öztürk Uludag University, Mechanical Eng. Department,

More information

Chapter 9: Genetic Algorithms

Chapter 9: Genetic Algorithms Computational Intelligence: Second Edition Contents Compact Overview First proposed by Fraser in 1957 Later by Bremermann in 1962 and Reed et al in 1967 Popularized by Holland in 1975 Genetic algorithms

More information

International Conference on Computer Applications in Shipbuilding (ICCAS-2009) Shanghai, China Vol.2, pp

International Conference on Computer Applications in Shipbuilding (ICCAS-2009) Shanghai, China Vol.2, pp AUTOMATIC DESIGN FOR PIPE ARRANGEMENT CONSIDERING VALVE OPERATIONALITY H Kimura, Kyushu University, Japan S Iehira, Kyushu University, Japan SUMMARY We propose a novel evaluation method of valve operationality

More information

Machine Evolution. Machine Evolution. Let s look at. Machine Evolution. Machine Evolution. Machine Evolution. Machine Evolution

Machine Evolution. Machine Evolution. Let s look at. Machine Evolution. Machine Evolution. Machine Evolution. Machine Evolution Let s look at As you will see later in this course, neural networks can learn, that is, adapt to given constraints. For example, NNs can approximate a given function. In biology, such learning corresponds

More information

Constraint-Driven Floorplanning based on Genetic Algorithm

Constraint-Driven Floorplanning based on Genetic Algorithm Proceedings of the 2007 WSEAS International Conference on Computer Engineering and Applications, Gold Coast, Australia, January 17-19, 2007 147 Constraint-Driven Floorplanning based on Genetic Algorithm

More information

Evolutionary Linkage Creation between Information Sources in P2P Networks

Evolutionary Linkage Creation between Information Sources in P2P Networks Noname manuscript No. (will be inserted by the editor) Evolutionary Linkage Creation between Information Sources in P2P Networks Kei Ohnishi Mario Köppen Kaori Yoshida Received: date / Accepted: date Abstract

More information

Genetic Programming of Autonomous Agents. Functional Description and Complete System Block Diagram. Scott O'Dell

Genetic Programming of Autonomous Agents. Functional Description and Complete System Block Diagram. Scott O'Dell Genetic Programming of Autonomous Agents Functional Description and Complete System Block Diagram Scott O'Dell Advisors: Dr. Joel Schipper and Dr. Arnold Patton October 19, 2010 Introduction to Genetic

More information

ADAPTATION OF REPRESENTATION IN GP

ADAPTATION OF REPRESENTATION IN GP 1 ADAPTATION OF REPRESENTATION IN GP CEZARY Z. JANIKOW University of Missouri St. Louis Department of Mathematics and Computer Science St Louis, Missouri RAHUL A DESHPANDE University of Missouri St. Louis

More information

Genetic Programming of Autonomous Agents. Functional Requirements List and Performance Specifi cations. Scott O'Dell

Genetic Programming of Autonomous Agents. Functional Requirements List and Performance Specifi cations. Scott O'Dell Genetic Programming of Autonomous Agents Functional Requirements List and Performance Specifi cations Scott O'Dell Advisors: Dr. Joel Schipper and Dr. Arnold Patton November 23, 2010 GPAA 1 Project Goals

More information

A Kind of Wireless Sensor Network Coverage Optimization Algorithm Based on Genetic PSO

A Kind of Wireless Sensor Network Coverage Optimization Algorithm Based on Genetic PSO Sensors & Transducers 2013 by IFSA http://www.sensorsportal.com A Kind of Wireless Sensor Network Coverage Optimization Algorithm Based on Genetic PSO Yinghui HUANG School of Electronics and Information,

More information

Using Genetic Algorithms to Solve the Box Stacking Problem

Using Genetic Algorithms to Solve the Box Stacking Problem Using Genetic Algorithms to Solve the Box Stacking Problem Jenniffer Estrada, Kris Lee, Ryan Edgar October 7th, 2010 Abstract The box stacking or strip stacking problem is exceedingly difficult to solve

More information

An Evolutionary Algorithm for the Multi-objective Shortest Path Problem

An Evolutionary Algorithm for the Multi-objective Shortest Path Problem An Evolutionary Algorithm for the Multi-objective Shortest Path Problem Fangguo He Huan Qi Qiong Fan Institute of Systems Engineering, Huazhong University of Science & Technology, Wuhan 430074, P. R. China

More information

CHAPTER 5 ENERGY MANAGEMENT USING FUZZY GENETIC APPROACH IN WSN

CHAPTER 5 ENERGY MANAGEMENT USING FUZZY GENETIC APPROACH IN WSN 97 CHAPTER 5 ENERGY MANAGEMENT USING FUZZY GENETIC APPROACH IN WSN 5.1 INTRODUCTION Fuzzy systems have been applied to the area of routing in ad hoc networks, aiming to obtain more adaptive and flexible

More information

Adaptive Crossover in Genetic Algorithms Using Statistics Mechanism

Adaptive Crossover in Genetic Algorithms Using Statistics Mechanism in Artificial Life VIII, Standish, Abbass, Bedau (eds)(mit Press) 2002. pp 182 185 1 Adaptive Crossover in Genetic Algorithms Using Statistics Mechanism Shengxiang Yang Department of Mathematics and Computer

More information

Mutations for Permutations

Mutations for Permutations Mutations for Permutations Insert mutation: Pick two allele values at random Move the second to follow the first, shifting the rest along to accommodate Note: this preserves most of the order and adjacency

More information

ARTIFICIAL INTELLIGENCE (CSCU9YE ) LECTURE 5: EVOLUTIONARY ALGORITHMS

ARTIFICIAL INTELLIGENCE (CSCU9YE ) LECTURE 5: EVOLUTIONARY ALGORITHMS ARTIFICIAL INTELLIGENCE (CSCU9YE ) LECTURE 5: EVOLUTIONARY ALGORITHMS Gabriela Ochoa http://www.cs.stir.ac.uk/~goc/ OUTLINE Optimisation problems Optimisation & search Two Examples The knapsack problem

More information

Literature Review On Implementing Binary Knapsack problem

Literature Review On Implementing Binary Knapsack problem Literature Review On Implementing Binary Knapsack problem Ms. Niyati Raj, Prof. Jahnavi Vitthalpura PG student Department of Information Technology, L.D. College of Engineering, Ahmedabad, India Assistant

More information

ADAPTIVE TILE CODING METHODS FOR THE GENERALIZATION OF VALUE FUNCTIONS IN THE RL STATE SPACE A THESIS SUBMITTED TO THE FACULTY OF THE GRADUATE SCHOOL

ADAPTIVE TILE CODING METHODS FOR THE GENERALIZATION OF VALUE FUNCTIONS IN THE RL STATE SPACE A THESIS SUBMITTED TO THE FACULTY OF THE GRADUATE SCHOOL ADAPTIVE TILE CODING METHODS FOR THE GENERALIZATION OF VALUE FUNCTIONS IN THE RL STATE SPACE A THESIS SUBMITTED TO THE FACULTY OF THE GRADUATE SCHOOL OF THE UNIVERSITY OF MINNESOTA BY BHARAT SIGINAM IN

More information

Applied Cloning Techniques for a Genetic Algorithm Used in Evolvable Hardware Design

Applied Cloning Techniques for a Genetic Algorithm Used in Evolvable Hardware Design Applied Cloning Techniques for a Genetic Algorithm Used in Evolvable Hardware Design Viet C. Trinh vtrinh@isl.ucf.edu Gregory A. Holifield greg.holifield@us.army.mil School of Electrical Engineering and

More information

Revision of a Floating-Point Genetic Algorithm GENOCOP V for Nonlinear Programming Problems

Revision of a Floating-Point Genetic Algorithm GENOCOP V for Nonlinear Programming Problems 4 The Open Cybernetics and Systemics Journal, 008,, 4-9 Revision of a Floating-Point Genetic Algorithm GENOCOP V for Nonlinear Programming Problems K. Kato *, M. Sakawa and H. Katagiri Department of Artificial

More information

Learning Composite Operators for Object Detection

Learning Composite Operators for Object Detection Real-World Applications Learning Composite Operators for Object Detection Bir Bhanu and Yingqiang Lin Center for Research in Intelligent Systems University of California, Riverside, CA, 92521, USA Email:

More information

Review: Final Exam CPSC Artificial Intelligence Michael M. Richter

Review: Final Exam CPSC Artificial Intelligence Michael M. Richter Review: Final Exam Model for a Learning Step Learner initially Environm ent Teacher Compare s pe c ia l Information Control Correct Learning criteria Feedback changed Learner after Learning Learning by

More information

The Genetic Algorithm for finding the maxima of single-variable functions

The Genetic Algorithm for finding the maxima of single-variable functions Research Inventy: International Journal Of Engineering And Science Vol.4, Issue 3(March 2014), PP 46-54 Issn (e): 2278-4721, Issn (p):2319-6483, www.researchinventy.com The Genetic Algorithm for finding

More information

Genetic Programming: A study on Computer Language

Genetic Programming: A study on Computer Language Genetic Programming: A study on Computer Language Nilam Choudhary Prof.(Dr.) Baldev Singh Er. Gaurav Bagaria Abstract- this paper describes genetic programming in more depth, assuming that the reader is

More information

A New Selection Operator - CSM in Genetic Algorithms for Solving the TSP

A New Selection Operator - CSM in Genetic Algorithms for Solving the TSP A New Selection Operator - CSM in Genetic Algorithms for Solving the TSP Wael Raef Alkhayri Fahed Al duwairi High School Aljabereyah, Kuwait Suhail Sami Owais Applied Science Private University Amman,

More information

Automata Construct with Genetic Algorithm

Automata Construct with Genetic Algorithm Automata Construct with Genetic Algorithm Vít Fábera Department of Informatics and Telecommunication, Faculty of Transportation Sciences, Czech Technical University, Konviktská 2, Praha, Czech Republic,

More information

GENETIC NETWORK PROGRAMMING-REINFORCEMENT LEARNING BASED SAFE AND SMOOTH MOBILE ROBOT NAVIGATION IN UNKNOWN DYNAMIC ENVIRONMENTS

GENETIC NETWORK PROGRAMMING-REINFORCEMENT LEARNING BASED SAFE AND SMOOTH MOBILE ROBOT NAVIGATION IN UNKNOWN DYNAMIC ENVIRONMENTS GENETIC NETWORK PROGRAMMING-REINFORCEMENT LEARNING BASED SAFE AND SMOOTH MOBILE ROBOT NAVIGATION IN UNKNOWN DYNAMIC ENVIRONMENTS 1,2 AHMED H. M. FINDI, 3 MOHAMMAD H. MARHABAN, 4 RAJA KAMIL, 5 MOHD KHAIR

More information

Experimental Comparison of Different Techniques to Generate Adaptive Sequences

Experimental Comparison of Different Techniques to Generate Adaptive Sequences Experimental Comparison of Different Techniques to Generate Adaptive Sequences Carlos Molinero 1, Manuel Núñez 1 and Robert M. Hierons 2 1 Departamento de Sistemas Informáticos y Computación, Universidad

More information

Ensemble Image Classification Method Based on Genetic Image Network

Ensemble Image Classification Method Based on Genetic Image Network Ensemble Image Classification Method Based on Genetic Image Network Shiro Nakayama, Shinichi Shirakawa, Noriko Yata and Tomoharu Nagao Graduate School of Environment and Information Sciences, Yokohama

More information

Genetic Algorithms for Vision and Pattern Recognition

Genetic Algorithms for Vision and Pattern Recognition Genetic Algorithms for Vision and Pattern Recognition Faiz Ul Wahab 11/8/2014 1 Objective To solve for optimization of computer vision problems using genetic algorithms 11/8/2014 2 Timeline Problem: Computer

More information

CHAPTER 2 CONVENTIONAL AND NON-CONVENTIONAL TECHNIQUES TO SOLVE ORPD PROBLEM

CHAPTER 2 CONVENTIONAL AND NON-CONVENTIONAL TECHNIQUES TO SOLVE ORPD PROBLEM 20 CHAPTER 2 CONVENTIONAL AND NON-CONVENTIONAL TECHNIQUES TO SOLVE ORPD PROBLEM 2.1 CLASSIFICATION OF CONVENTIONAL TECHNIQUES Classical optimization methods can be classified into two distinct groups:

More information

Computational Intelligence

Computational Intelligence Computational Intelligence Module 6 Evolutionary Computation Ajith Abraham Ph.D. Q What is the most powerful problem solver in the Universe? ΑThe (human) brain that created the wheel, New York, wars and

More information

CHAPTER 6 REAL-VALUED GENETIC ALGORITHMS

CHAPTER 6 REAL-VALUED GENETIC ALGORITHMS CHAPTER 6 REAL-VALUED GENETIC ALGORITHMS 6.1 Introduction Gradient-based algorithms have some weaknesses relative to engineering optimization. Specifically, it is difficult to use gradient-based algorithms

More information

Reinforcement Learning (2)

Reinforcement Learning (2) Reinforcement Learning (2) Bruno Bouzy 1 october 2013 This document is the second part of the «Reinforcement Learning» chapter of the «Agent oriented learning» teaching unit of the Master MI computer course.

More information

Santa Fe Trail Problem Solution Using Grammatical Evolution

Santa Fe Trail Problem Solution Using Grammatical Evolution 2012 International Conference on Industrial and Intelligent Information (ICIII 2012) IPCSIT vol.31 (2012) (2012) IACSIT Press, Singapore Santa Fe Trail Problem Solution Using Grammatical Evolution Hideyuki

More information

Genetic algorithms and finite element coupling for mechanical optimization

Genetic algorithms and finite element coupling for mechanical optimization Computer Aided Optimum Design in Engineering X 87 Genetic algorithms and finite element coupling for mechanical optimization G. Corriveau, R. Guilbault & A. Tahan Department of Mechanical Engineering,

More information

Genetic Image Network for Image Classification

Genetic Image Network for Image Classification Genetic Image Network for Image Classification Shinichi Shirakawa, Shiro Nakayama, and Tomoharu Nagao Graduate School of Environment and Information Sciences, Yokohama National University, 79-7, Tokiwadai,

More information

Artificial Intelligence

Artificial Intelligence Artificial Intelligence Informed Search and Exploration Chapter 4 (4.3 4.6) Searching: So Far We ve discussed how to build goal-based and utility-based agents that search to solve problems We ve also presented

More information

Optimization of Association Rule Mining through Genetic Algorithm

Optimization of Association Rule Mining through Genetic Algorithm Optimization of Association Rule Mining through Genetic Algorithm RUPALI HALDULAKAR School of Information Technology, Rajiv Gandhi Proudyogiki Vishwavidyalaya Bhopal, Madhya Pradesh India Prof. JITENDRA

More information

CHAPTER 6 HYBRID AI BASED IMAGE CLASSIFICATION TECHNIQUES

CHAPTER 6 HYBRID AI BASED IMAGE CLASSIFICATION TECHNIQUES CHAPTER 6 HYBRID AI BASED IMAGE CLASSIFICATION TECHNIQUES 6.1 INTRODUCTION The exploration of applications of ANN for image classification has yielded satisfactory results. But, the scope for improving

More information

International Journal of Digital Application & Contemporary research Website: (Volume 1, Issue 7, February 2013)

International Journal of Digital Application & Contemporary research Website:   (Volume 1, Issue 7, February 2013) Performance Analysis of GA and PSO over Economic Load Dispatch Problem Sakshi Rajpoot sakshirajpoot1988@gmail.com Dr. Sandeep Bhongade sandeepbhongade@rediffmail.com Abstract Economic Load dispatch problem

More information

Simultaneous Optimization of a Wheeled Mobile Robot Structure and a Control Parameter

Simultaneous Optimization of a Wheeled Mobile Robot Structure and a Control Parameter Fifth International Workshop on Computational Intelligence & Applications IEEE SMC Hiroshima Chapter, Hiroshima University, Japan, November 10, 11 & 12, 2009 Simultaneous Optimization of a Wheeled Mobile

More information

GENETIC ALGORITHM with Hands-On exercise

GENETIC ALGORITHM with Hands-On exercise GENETIC ALGORITHM with Hands-On exercise Adopted From Lecture by Michael Negnevitsky, Electrical Engineering & Computer Science University of Tasmania 1 Objective To understand the processes ie. GAs Basic

More information

Introduction to Genetic Algorithms. Based on Chapter 10 of Marsland Chapter 9 of Mitchell

Introduction to Genetic Algorithms. Based on Chapter 10 of Marsland Chapter 9 of Mitchell Introduction to Genetic Algorithms Based on Chapter 10 of Marsland Chapter 9 of Mitchell Genetic Algorithms - History Pioneered by John Holland in the 1970s Became popular in the late 1980s Based on ideas

More information

Escaping Local Optima: Genetic Algorithm

Escaping Local Optima: Genetic Algorithm Artificial Intelligence Escaping Local Optima: Genetic Algorithm Dae-Won Kim School of Computer Science & Engineering Chung-Ang University We re trying to escape local optima To achieve this, we have learned

More information

Homework 2: Search and Optimization

Homework 2: Search and Optimization Scott Chow ROB 537: Learning Based Control October 16, 2017 Homework 2: Search and Optimization 1 Introduction The Traveling Salesman Problem is a well-explored problem that has been shown to be NP-Complete.

More information

Evolving SQL Queries for Data Mining

Evolving SQL Queries for Data Mining Evolving SQL Queries for Data Mining Majid Salim and Xin Yao School of Computer Science, The University of Birmingham Edgbaston, Birmingham B15 2TT, UK {msc30mms,x.yao}@cs.bham.ac.uk Abstract. This paper

More information

Global Optimal Analysis of Variant Genetic Operations in Solar Tracking

Global Optimal Analysis of Variant Genetic Operations in Solar Tracking Australian Journal of Basic and Applied Sciences, 6(6): 6-14, 2012 ISSN 1991-8178 Global Optimal Analysis of Variant Genetic Operations in Solar Tracking D.F.Fam, S.P. Koh, S.K. Tiong, K.H.Chong Department

More information

Research Article Path Planning Using a Hybrid Evolutionary Algorithm Based on Tree Structure Encoding

Research Article Path Planning Using a Hybrid Evolutionary Algorithm Based on Tree Structure Encoding e Scientific World Journal, Article ID 746260, 8 pages http://dx.doi.org/10.1155/2014/746260 Research Article Path Planning Using a Hybrid Evolutionary Algorithm Based on Tree Structure Encoding Ming-Yi

More information

Automated Test Data Generation and Optimization Scheme Using Genetic Algorithm

Automated Test Data Generation and Optimization Scheme Using Genetic Algorithm 2011 International Conference on Software and Computer Applications IPCSIT vol.9 (2011) (2011) IACSIT Press, Singapore Automated Test Data Generation and Optimization Scheme Using Genetic Algorithm Roshni

More information

Improving Prediction Power in Simulation of Brassiere-Wearing Figures

Improving Prediction Power in Simulation of Brassiere-Wearing Figures Improving Prediction Power in Simulation of Brassiere-Wearing Figures Choi Dong-Eun*, Nakamura Kensuke** and Kurokawa Takao** * Venture Laboratory, Kyoto Institute of Technology **Graduate School of Science

More information

Ant colony optimization with genetic operations

Ant colony optimization with genetic operations Automation, Control and Intelligent Systems ; (): - Published online June, (http://www.sciencepublishinggroup.com/j/acis) doi:./j.acis.. Ant colony optimization with genetic operations Matej Ciba, Ivan

More information

Chapter 5 Components for Evolution of Modular Artificial Neural Networks

Chapter 5 Components for Evolution of Modular Artificial Neural Networks Chapter 5 Components for Evolution of Modular Artificial Neural Networks 5.1 Introduction In this chapter, the methods and components used for modular evolution of Artificial Neural Networks (ANNs) are

More information

Marco Wiering Intelligent Systems Group Utrecht University

Marco Wiering Intelligent Systems Group Utrecht University Reinforcement Learning for Robot Control Marco Wiering Intelligent Systems Group Utrecht University marco@cs.uu.nl 22-11-2004 Introduction Robots move in the physical environment to perform tasks The environment

More information

COMP SCI 5401 FS Iterated Prisoner s Dilemma: A Coevolutionary Genetic Programming Approach

COMP SCI 5401 FS Iterated Prisoner s Dilemma: A Coevolutionary Genetic Programming Approach COMP SCI 5401 FS2017 - Iterated Prisoner s Dilemma: A Coevolutionary Genetic Programming Approach Daniel Tauritz, Ph.D. November 17, 2017 Synopsis The goal of this assignment set is for you to become familiarized

More information

A TSK-Type Recurrent Fuzzy Network for Dynamic Systems Processing by Neural Network and Genetic Algorithms

A TSK-Type Recurrent Fuzzy Network for Dynamic Systems Processing by Neural Network and Genetic Algorithms IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 10, NO. 2, APRIL 2002 155 A TSK-Type Recurrent Fuzzy Network for Dynamic Systems Processing by Neural Network and Genetic Algorithms Chia-Feng Juang, Member, IEEE

More information

Using Genetic Programming to Evolve Robot Behaviours

Using Genetic Programming to Evolve Robot Behaviours Proceedings of the 3rd British Conference on Autonomous Mobile Robotics & Autonomous Systems, Manchester, 5th April 2001 Using Genetic Programming to Evolve Robot Behaviours Christopher Lazarus and Huosheng

More information

Genetic Algorithms. Kang Zheng Karl Schober

Genetic Algorithms. Kang Zheng Karl Schober Genetic Algorithms Kang Zheng Karl Schober Genetic algorithm What is Genetic algorithm? A genetic algorithm (or GA) is a search technique used in computing to find true or approximate solutions to optimization

More information

Introduction to Optimization

Introduction to Optimization Introduction to Optimization Approximation Algorithms and Heuristics November 21, 2016 École Centrale Paris, Châtenay-Malabry, France Dimo Brockhoff Inria Saclay Ile-de-France 2 Exercise: The Knapsack

More information

Genetic Algorithms. PHY 604: Computational Methods in Physics and Astrophysics II

Genetic Algorithms. PHY 604: Computational Methods in Physics and Astrophysics II Genetic Algorithms Genetic Algorithms Iterative method for doing optimization Inspiration from biology General idea (see Pang or Wikipedia for more details): Create a collection of organisms/individuals

More information

3.6.2 Generating admissible heuristics from relaxed problems

3.6.2 Generating admissible heuristics from relaxed problems 3.6.2 Generating admissible heuristics from relaxed problems To come up with heuristic functions one can study relaxed problems from which some restrictions of the original problem have been removed The

More information

Introduction to Optimization

Introduction to Optimization Introduction to Optimization Approximation Algorithms and Heuristics November 6, 2015 École Centrale Paris, Châtenay-Malabry, France Dimo Brockhoff INRIA Lille Nord Europe 2 Exercise: The Knapsack Problem

More information

1 Lab 5: Particle Swarm Optimization

1 Lab 5: Particle Swarm Optimization 1 Lab 5: Particle Swarm Optimization This laboratory requires the following: (The development tools are installed in GR B0 01 already): C development tools (gcc, make, etc.) Webots simulation software

More information

Path Planning for a Robot Manipulator based on Probabilistic Roadmap and Reinforcement Learning

Path Planning for a Robot Manipulator based on Probabilistic Roadmap and Reinforcement Learning 674 International Journal Jung-Jun of Control, Park, Automation, Ji-Hun Kim, and and Systems, Jae-Bok vol. Song 5, no. 6, pp. 674-680, December 2007 Path Planning for a Robot Manipulator based on Probabilistic

More information

Evolutionary Computation Algorithms for Cryptanalysis: A Study

Evolutionary Computation Algorithms for Cryptanalysis: A Study Evolutionary Computation Algorithms for Cryptanalysis: A Study Poonam Garg Information Technology and Management Dept. Institute of Management Technology Ghaziabad, India pgarg@imt.edu Abstract The cryptanalysis

More information

A Fuzzy Reinforcement Learning for a Ball Interception Problem

A Fuzzy Reinforcement Learning for a Ball Interception Problem A Fuzzy Reinforcement Learning for a Ball Interception Problem Tomoharu Nakashima, Masayo Udo, and Hisao Ishibuchi Department of Industrial Engineering, Osaka Prefecture University Gakuen-cho 1-1, Sakai,

More information

THE EFFECT OF SEGREGATION IN NON- REPEATED PRISONER'S DILEMMA

THE EFFECT OF SEGREGATION IN NON- REPEATED PRISONER'S DILEMMA THE EFFECT OF SEGREGATION IN NON- REPEATED PRISONER'S DILEMMA Thomas Nordli University of South-Eastern Norway, Norway ABSTRACT This article consolidates the idea that non-random pairing can promote the

More information

Comparison of Evolutionary Multiobjective Optimization with Reference Solution-Based Single-Objective Approach

Comparison of Evolutionary Multiobjective Optimization with Reference Solution-Based Single-Objective Approach Comparison of Evolutionary Multiobjective Optimization with Reference Solution-Based Single-Objective Approach Hisao Ishibuchi Graduate School of Engineering Osaka Prefecture University Sakai, Osaka 599-853,

More information

Clustering Analysis of Simple K Means Algorithm for Various Data Sets in Function Optimization Problem (Fop) of Evolutionary Programming

Clustering Analysis of Simple K Means Algorithm for Various Data Sets in Function Optimization Problem (Fop) of Evolutionary Programming Clustering Analysis of Simple K Means Algorithm for Various Data Sets in Function Optimization Problem (Fop) of Evolutionary Programming R. Karthick 1, Dr. Malathi.A 2 Research Scholar, Department of Computer

More information

Evolutionary Computation. Chao Lan

Evolutionary Computation. Chao Lan Evolutionary Computation Chao Lan Outline Introduction Genetic Algorithm Evolutionary Strategy Genetic Programming Introduction Evolutionary strategy can jointly optimize multiple variables. - e.g., max

More information

Boolean network robotics

Boolean network robotics Boolean network robotics An example of ongoing robotics research Andrea Roli andrea.roli@unibo.it DISI - Dept. of Computer Science and Engineering Campus of Cesena Alma Mater Studiorum Università di Bologna

More information

The k-means Algorithm and Genetic Algorithm

The k-means Algorithm and Genetic Algorithm The k-means Algorithm and Genetic Algorithm k-means algorithm Genetic algorithm Rough set approach Fuzzy set approaches Chapter 8 2 The K-Means Algorithm The K-Means algorithm is a simple yet effective

More information

CHAPTER 4 GENETIC ALGORITHM

CHAPTER 4 GENETIC ALGORITHM 69 CHAPTER 4 GENETIC ALGORITHM 4.1 INTRODUCTION Genetic Algorithms (GAs) were first proposed by John Holland (Holland 1975) whose ideas were applied and expanded on by Goldberg (Goldberg 1989). GAs is

More information

A Method Based Genetic Algorithm for Pipe Routing Design

A Method Based Genetic Algorithm for Pipe Routing Design 5th International Conference on Advanced Engineering Materials and Technology (AEMT 2015) A Method Based Genetic Algorithm for Pipe Routing Design Changtao Wang 1, a, Xiaotong Sun 2,b,Tiancheng Yuan 3,c

More information

A GENETIC ALGORITHM FOR CLUSTERING ON VERY LARGE DATA SETS

A GENETIC ALGORITHM FOR CLUSTERING ON VERY LARGE DATA SETS A GENETIC ALGORITHM FOR CLUSTERING ON VERY LARGE DATA SETS Jim Gasvoda and Qin Ding Department of Computer Science, Pennsylvania State University at Harrisburg, Middletown, PA 17057, USA {jmg289, qding}@psu.edu

More information

Applying genetic algorithm on power system stabilizer for stabilization of power system

Applying genetic algorithm on power system stabilizer for stabilization of power system Applying genetic algorithm on power system stabilizer for stabilization of power system 1,3 Arnawan Hasibuan and 2,3 Syafrudin 1 Engineering Department of Malikussaleh University, Lhokseumawe, Indonesia;

More information

Using Genetic Algorithms in Integer Programming for Decision Support

Using Genetic Algorithms in Integer Programming for Decision Support Doi:10.5901/ajis.2014.v3n6p11 Abstract Using Genetic Algorithms in Integer Programming for Decision Support Dr. Youcef Souar Omar Mouffok Taher Moulay University Saida, Algeria Email:Syoucef12@yahoo.fr

More information

Gauss-Sigmoid Neural Network

Gauss-Sigmoid Neural Network Gauss-Sigmoid Neural Network Katsunari SHIBATA and Koji ITO Tokyo Institute of Technology, Yokohama, JAPAN shibata@ito.dis.titech.ac.jp Abstract- Recently RBF(Radial Basis Function)-based networks have

More information

Solving Traveling Salesman Problem Using Parallel Genetic. Algorithm and Simulated Annealing

Solving Traveling Salesman Problem Using Parallel Genetic. Algorithm and Simulated Annealing Solving Traveling Salesman Problem Using Parallel Genetic Algorithm and Simulated Annealing Fan Yang May 18, 2010 Abstract The traveling salesman problem (TSP) is to find a tour of a given number of cities

More information

Previous Lecture Genetic Programming

Previous Lecture Genetic Programming Genetic Programming Previous Lecture Constraint Handling Penalty Approach Penalize fitness for infeasible solutions, depending on distance from feasible region Balanace between under- and over-penalization

More information

Radio Network Planning with Combinatorial Optimisation Algorithms

Radio Network Planning with Combinatorial Optimisation Algorithms Author manuscript, published in "ACTS Mobile Telecommunications Summit 96, Granada : Spain (1996)" Radio Network Planning with Combinatorial Optimisation Algorithms P. Calégari, F. Guidec, P. Kuonen, EPFL,

More information

GENETIC ALGORITHM METHOD FOR COMPUTER AIDED QUALITY CONTROL

GENETIC ALGORITHM METHOD FOR COMPUTER AIDED QUALITY CONTROL 3 rd Research/Expert Conference with International Participations QUALITY 2003, Zenica, B&H, 13 and 14 November, 2003 GENETIC ALGORITHM METHOD FOR COMPUTER AIDED QUALITY CONTROL Miha Kovacic, Miran Brezocnik

More information

INTERACTIVE MULTI-OBJECTIVE GENETIC ALGORITHMS FOR THE BUS DRIVER SCHEDULING PROBLEM

INTERACTIVE MULTI-OBJECTIVE GENETIC ALGORITHMS FOR THE BUS DRIVER SCHEDULING PROBLEM Advanced OR and AI Methods in Transportation INTERACTIVE MULTI-OBJECTIVE GENETIC ALGORITHMS FOR THE BUS DRIVER SCHEDULING PROBLEM Jorge PINHO DE SOUSA 1, Teresa GALVÃO DIAS 1, João FALCÃO E CUNHA 1 Abstract.

More information

Offspring Generation Method using Delaunay Triangulation for Real-Coded Genetic Algorithms

Offspring Generation Method using Delaunay Triangulation for Real-Coded Genetic Algorithms Offspring Generation Method using Delaunay Triangulation for Real-Coded Genetic Algorithms Hisashi Shimosaka 1, Tomoyuki Hiroyasu 2, and Mitsunori Miki 2 1 Graduate School of Engineering, Doshisha University,

More information

Genetic Programming Part 1

Genetic Programming Part 1 Genetic Programming Part 1 Evolutionary Computation Lecture 11 Thorsten Schnier 06/11/2009 Previous Lecture Multi-objective Optimization Pareto optimality Hyper-volume based indicators Recent lectures

More information

How Santa Fe Ants Evolve

How Santa Fe Ants Evolve How Santa Fe Ants Evolve Dominic Wilson, Devinder Kaur, Abstract The Santa Fe Ant model problem has been extensively used to investigate, test and evaluate evolutionary computing systems and methods over

More information

Variations on Genetic Cellular Automata

Variations on Genetic Cellular Automata Variations on Genetic Cellular Automata Alice Durand David Olson Physics Department amdurand@ucdavis.edu daolson@ucdavis.edu Abstract: We investigated the properties of cellular automata with three or

More information

The movement of the dimmer firefly i towards the brighter firefly j in terms of the dimmer one s updated location is determined by the following equat

The movement of the dimmer firefly i towards the brighter firefly j in terms of the dimmer one s updated location is determined by the following equat An Improved Firefly Algorithm for Optimization Problems Amarita Ritthipakdee 1, Arit Thammano, Nol Premasathian 3, and Bunyarit Uyyanonvara 4 Abstract Optimization problem is one of the most difficult

More information

Immune Optimization Design of Diesel Engine Valve Spring Based on the Artificial Fish Swarm

Immune Optimization Design of Diesel Engine Valve Spring Based on the Artificial Fish Swarm IOSR Journal of Computer Engineering (IOSR-JCE) e-issn: 2278-661, p- ISSN: 2278-8727Volume 16, Issue 4, Ver. II (Jul-Aug. 214), PP 54-59 Immune Optimization Design of Diesel Engine Valve Spring Based on

More information

Neural Network Weight Selection Using Genetic Algorithms

Neural Network Weight Selection Using Genetic Algorithms Neural Network Weight Selection Using Genetic Algorithms David Montana presented by: Carl Fink, Hongyi Chen, Jack Cheng, Xinglong Li, Bruce Lin, Chongjie Zhang April 12, 2005 1 Neural Networks Neural networks

More information

4/22/2014. Genetic Algorithms. Diwakar Yagyasen Department of Computer Science BBDNITM. Introduction

4/22/2014. Genetic Algorithms. Diwakar Yagyasen Department of Computer Science BBDNITM. Introduction 4/22/24 s Diwakar Yagyasen Department of Computer Science BBDNITM Visit dylycknow.weebly.com for detail 2 The basic purpose of a genetic algorithm () is to mimic Nature s evolutionary approach The algorithm

More information

1 Lab + Hwk 5: Particle Swarm Optimization

1 Lab + Hwk 5: Particle Swarm Optimization 1 Lab + Hwk 5: Particle Swarm Optimization This laboratory requires the following equipment: C programming tools (gcc, make), already installed in GR B001 Webots simulation software Webots User Guide Webots

More information