混合式贝叶斯网络结构增量学习算法研究-控制与决策.pdf

绗?5 鍗 绗? 鏈 鎺 鍒 涓 鍐 绛 2010 骞 6 鏈 Vol. 25 No. 6 Control and Decision Jun. 2010 鏂囩珷缂栧彿: 1001-0920 (2010) 06-0925-04 娣峰悎寮忚礉鍙舵柉缃戠粶缁撴瀯澧為噺瀛︿範绠楁硶鐮旂┒ 鍙 杈? 璋皯鍗 (鍖椾含澶у a. 淇℃伅绉戝鎶€鏈闄紝b. 瑙嗚涓庡惉瑙変俊鎭鐞嗘暀鑲查儴閲嶇偣瀹為獙瀹わ紝鍖椾含100871) 鎽 瑕? 鎻愬嚭涓€绉嶆贩鍚堝紡璐濆彾鏂綉缁滅粨鏋勫閲忓涔犵畻娉? 棣栧厛鎻愬嚭澶氶」寮忔椂闂寸殑闄愬埗鎬у涔犳妧鏈? 涓烘瘡涓彉閲忓缓绔 鍊欓€夌埗鑺傜偣闆嗗悎; 鐒跺悗, 渚濇嵁鍊欓€夌埗鑺傜偣闆嗗悎, 鍒╃敤鎼滅储鎶€鏈褰撳墠缃戠粶杩涜澧為噺瀛︿範. 璇ョ畻娉曠殑澶嶆潅搴︽樉钁椾綆浜庣洰 鍓嶆渶浼樼殑璐濆彾鏂綉缁滃閲忓涔犵畻娉? 鐞嗚涓庡疄楠屽潎琛ㄦ槑, 鎵€澶勭悊鐨勯棶棰樿秺澶嶆潅, 璇ョ畻娉曞湪璁＄畻澶嶆潅搴︽柟闈㈢殑浼樺娍瓒婃槑 鏄? 鍏抽敭璇岤 璐濆彾鏂綉缁滐紱澧為噺瀛︿範锛涚粨鏋勫涔 涓浘鍒嗙被鍙耳 TP181 鏂囩尞鏍囪瘑鐮? A A new hybrid incremental learning algorithm for Bayesian network structures SHI Da , TAN Shao-hua (a. School of Electronic Engineering and Computer Science锛宐. The Key Laboratory of Machine Perception of Ministry of Education 锛孭eking University 锛孊eijing 100871 锛孋hina. Correspondent: SHI Da锛孍-mail: lanceconan@) Abstract: A hybrid incremental learning algorithm for Bayesian network structures is proposed. A polynomial time constraint-based technique is developed to build up a candidate parent-set for each variable, and then a search-and-score technique is employed to re铿乶e the current network structure under the guidance of those candidate parent-sets. Experiment results show that the algorithm offers considerable computational savings while always obtaining slightly better results on model accuracy compared to the existing incremental algorithms. The more complex the real world problems are, the more signi铿乧ant the advantage on computational complexity the algorithm keeps over the existing incremental algorithms is. Key words: Bayesian networks 锛汭ncremental learning 锛汼tructure learning 1 寮曞紩寮 瑷€瑷€瑷€