-
公开(公告)号:WO2005089340A3
公开(公告)日:2008-01-10
申请号:PCT/US2005008648
申请日:2005-03-15
Applicant: UNIV SOUTHERN CALIFORNIA , KNIGHT KEVIN , GRAEHL JONATHAN
Inventor: KNIGHT KEVIN , GRAEHL JONATHAN
CPC classification number: G06F17/2775 , G06F17/2818
Abstract: Training using tree transducers is described. Given sample input/output pairs as training (100, 110), and given a set of tree transducer rules (120), the information is combined to yield locally optimal weights for those rules (140). This combination is carried out by building a weighted derivation forest for each input/output pair and applying counting methods to those forests (130).
Abstract translation: 描述使用树形传感器的训练。 给定样本输入/输出对作为训练(100,110),并且给定一组树传感器规则(120),将该信息组合以产生用于那些规则的局部最优权重(140)。 这种组合是通过为每个输入/输出对构建加权推导森林,并对这些森林应用计数方法来执行的(130)。