A parallel algorithm for constrained two-staged two-dimensional cutting problems

In this paper, we solve the two-staged two-dimensional cutting problem using a parallel algorithm. The proposed approach combines two main features: beam search (BS) and strip generation solution procedures (SGSP). BS employs a truncated tree-search, where a selected subset of generated nodes are retuned for further search. SGSP, a constructive procedure, combines a (sub)set of strips for providing both partial lower and complementary upper bounds. The algorithm explores in parallel a subset of selected nodes following the master-slave paradigm. The master processor serves to guide the search-resolution and each slave processor develops its proper way, trying a global convergence. The aim of such an approach is to show how the parallelism is able to efficiently solve large-scale instances, by providing new solutions within a consistently reduced runtime. Extensive computational testing on instances, taken from the literature, shows the effectiveness of the proposed approach.     

基于柱搜索的高阶依存句法分析

该文提出使用所有的孙子节点构成祖孙特征的高阶依存模型,并且使用柱搜索策略限制搜索空间,最终找到近似最优依存树。另外,该文以较小的时间复杂度为代价,使用了丰富的依存关系特征,并且允许模型在解码的过程中进行依存关系选择。作者参加了CoNLL 2009年多语依存句法分析和语义角色标注国际评测,最终获得联合任务总成绩第一名,依存句法分析总成绩第三名。