Each　adaptive　agent　always　has　bounded　rationality　and　incomplete　knowledge　in　complex　adaptive　system.　And,　agents　with　self-learning　ability　can　modify　their　behaviors　constantly　to　evolve　adaptively　according　to　the　feedback　of　interaction.　To　address　this　issue,　a　memory　model　was　introduced　into　fictitious　play,　then　a　memory-weighted　learning　model　was　proposed　used　for　directing　the　strategy　selection　of　adaptive　agents,　and　finally　the　learning　convergence　in　different　types　of　games　was　observed　based　on　swarm　platform.　The　experience　results　show　that　although　the　limits　of　forgetting　and　local　interaction　exist,　the　agents　could　adjust　strategy　according　to　the　information　observed　continuously,　and　lead　to　the　realization　of　adaptive　learning.　©　2011　Springer-Verlag.