With　the　explosion　of　multimedia　applications,　the　network　QoS　is　facing　a　set　of　challenges　especially　in　congestion　control.　Active　queue　management　(AQM),　which　plays　an　important　role　in　network　congestion　control,　has　been　proved　necessary　for　decades.　Recently,　as　the　widespread　bufferbloat　being　exposed,　AQM　has　been　paid　more　and　more　attention.　However,　traditional　manually　designed　AQMs　still　exist　some　problems　especially　in　parameter-tuning　and　scenarios　adaption.　Instead　of　designing　a　perfect　AQM　for　all　scenarios,　which　is　nearly　impossible,　we　try　to　make　the　computer　generate　an　AQM　for　the　scenario　specified　by　users.　We＇ve　developed　a　program　called　Kemy　based　on　off-line　machine　learning　technologies.　The　Kemy-generated　AQM　is　evaluated　in　various　scenarios　and　achieves　the　goals　of　solving　bufferbloat　problem.　Compared　to　some　representative　human-designed　AQMs,　Kemy-generated　AQM　performs　even　better　in　some　cases.