Knowledge-based Extraction ofAreaofExpertise forCooperation inLearning
暂无分享,去创建一个
r masoud.asadpourgepfl.ch rsiegwartgethz.ch Abstract- Usingeachother's knowledge andexpertise in learning -whatwe callcooperation inlearning- isoneofthe majorexisting methods toreduce thenumberoflearning trials, whichisquite crucial forrealworldapplications. Insituated systems, robots becomeexpert indifferent areasduetobeing exposed todifferent situations andtasks. As a consequence, AreasOfExpertise (AOE)oftheotheragents mustbedetected before usingtheir knowledge, especially whentheexchanged knowledge isnotabstract, andsimple information exchange mightresult inincorrect knowledge, whichisthecaseforQ- learning agents. Inthispaperwe introduce anapproach forextraction of AOE ofagents forcooperation inlearning using their Q-tables. Theevaluating robot usesabehavioral measuretoevaluate itself, inordertofindasetofstates itisexpert in.Thatsetisused, then, alongwithaQ-table-based feature forextraction ofareas ofexpertise ofotherrobots bymeansofa classifier. Extracted areasaremergedinthelast stage. Theproposed methodistested bothinextensive simulations andinrealworldexperiments using mobile robots. Theresults showeffectiveness oftheintroduced approach, bothinaccurate extraction ofareasofexpertise andincreasing thequality ofthe combinedknowledge, evenwhen,thereareuncertainty and perceptual aliasing intheapplication andtherobot. IndexTerms- Cooperation inlearning, areaofexpertise, knowledge evaluation, Q-learning, Multi-robot learning.
[1] C. Boutilier,et al. Accelerating Reinforcement Learning through Implicit Imitation , 2003, J. Artif. Intell. Res..
[2] Majid Nili Ahmadabadi,et al. Expertness based cooperative Q-learning , 2002, IEEE Trans. Syst. Man Cybern. Part B.
[3] Thomas G. Dietterich. Hierarchical Reinforcement Learning with the MAXQ Value Function Decomposition , 1999, J. Artif. Intell. Res..