在中小学英语教学中,阅读理解测试的首个提问是至关重要的,以其作为考察切入点能够引导后续问题的提出,以便检验学生理解能力和逻辑思维能力。现有的提问生成方法产生的问句内容发散,难以达到教学考察目的,并且严重依赖基础知识库和计...在中小学英语教学中,阅读理解测试的首个提问是至关重要的,以其作为考察切入点能够引导后续问题的提出,以便检验学生理解能力和逻辑思维能力。现有的提问生成方法产生的问句内容发散,难以达到教学考察目的,并且严重依赖基础知识库和计算资源。为此,提出采用摘要和句法技术,以低算力高效率的方式实现首个提问生成。该方法从英语阅读正文中,选择核心摘要句以便聚焦正文主旨,并通过转译规则改写摘要语句的表述方式。对于转译后的核心摘要句,分析其主谓宾成分信息,确定关注对象,匹配相应的疑问词,进而根据语法规则生成首个提问。实验表明,提出的方法的语义相似度Bert Score和精确率BLRU-4分别达到了67.15和16.07,在句法、语义和可回答性方面均优于基线方法,能够有效生成适配于英语教学场景的首次提问。In basic English teaching, the first question in reading tests is crucial, which acts as an examination entry point leading to subsequent questions and testing students’ comprehension and logical thinking skills. Existing methods generate questions that are divergent in content, hardly meaningful for pedagogical investigation, and heavily dependent on knowledge base and computational resources. Therefore, this paper proposed a method using summarization and syntax to achieve first question generation in a low-computing and high-efficient manner. The method selected key summary sentence from the English reading text to focus on the main idea of the text, and changed its presentation by applying translation rules. For the paraphrased key summary sentence, its subject-predicate-object components was analyzed to determine the subject of concern, matched the corresponding question words, and then generated the first question according to the grammatical rules. Experimental results show that the semantic similarity Bert Score and accuracy BLRU-4 of proposed method reach 67.15 and 16.07, respectively, which outperforms the baseline method in terms of syntax, semantics, and answerability. Experiments also demonstrate that the proposed method can effectively generate first questions adapted to English teaching scenarios.展开更多
针对云计算系统在运行过程中由于计算节点空闲而产生大量空闲能耗,以及由于不匹配任务调度而产生大量"奢侈"能耗的能耗浪费问题,提出一种通过任务调度方式的能耗优化管理方法.首先,用排队模型对云计算系统进行建模,分析云计...针对云计算系统在运行过程中由于计算节点空闲而产生大量空闲能耗,以及由于不匹配任务调度而产生大量"奢侈"能耗的能耗浪费问题,提出一种通过任务调度方式的能耗优化管理方法.首先,用排队模型对云计算系统进行建模,分析云计算系统的平均响应时间和平均功率,建立云计算系统的能耗模型.然后提出基于大服务强度和小执行能耗的任务调度策略,分别针对空闲能耗和"奢侈"能耗进行优化控制.基于该调度策略,设计满足性能约束的最小期望执行能耗调度算法ME3PC(minimum expectation execution energy with performance constraints).实验结果表明,该算法在保证执行性能的前提下,可大幅度降低云计算系统的能耗开销.展开更多
矩阵主特征向量(principal eigenvectors computing,PEC)的求解是科学与工程计算中的一个重要问题。随着图形处理单元通用计算(general-purpose computing on graphics pro cessing unit,GPGPU)的兴起,利用GPU来优化大规模稀疏矩阵的图...矩阵主特征向量(principal eigenvectors computing,PEC)的求解是科学与工程计算中的一个重要问题。随着图形处理单元通用计算(general-purpose computing on graphics pro cessing unit,GPGPU)的兴起,利用GPU来优化大规模稀疏矩阵的图形处理单元求解得到了广泛关注。分别从应用特征和GPU体系结构特征两方面分析了PEC运算的性能瓶颈,提出了一种面向GPU的稀疏矩阵存储格式——GPU-ELL和一个针对GPU的线程优化映射策略,并设计了相应的PEC优化执行算法。在ATI HD Radeon5850上的实验结果表明,相对于传统CPU,该方案获得了最多200倍左右的加速,相对于已有GPU上的实现,也获得了2倍的加速。展开更多
文摘在中小学英语教学中,阅读理解测试的首个提问是至关重要的,以其作为考察切入点能够引导后续问题的提出,以便检验学生理解能力和逻辑思维能力。现有的提问生成方法产生的问句内容发散,难以达到教学考察目的,并且严重依赖基础知识库和计算资源。为此,提出采用摘要和句法技术,以低算力高效率的方式实现首个提问生成。该方法从英语阅读正文中,选择核心摘要句以便聚焦正文主旨,并通过转译规则改写摘要语句的表述方式。对于转译后的核心摘要句,分析其主谓宾成分信息,确定关注对象,匹配相应的疑问词,进而根据语法规则生成首个提问。实验表明,提出的方法的语义相似度Bert Score和精确率BLRU-4分别达到了67.15和16.07,在句法、语义和可回答性方面均优于基线方法,能够有效生成适配于英语教学场景的首次提问。In basic English teaching, the first question in reading tests is crucial, which acts as an examination entry point leading to subsequent questions and testing students’ comprehension and logical thinking skills. Existing methods generate questions that are divergent in content, hardly meaningful for pedagogical investigation, and heavily dependent on knowledge base and computational resources. Therefore, this paper proposed a method using summarization and syntax to achieve first question generation in a low-computing and high-efficient manner. The method selected key summary sentence from the English reading text to focus on the main idea of the text, and changed its presentation by applying translation rules. For the paraphrased key summary sentence, its subject-predicate-object components was analyzed to determine the subject of concern, matched the corresponding question words, and then generated the first question according to the grammatical rules. Experimental results show that the semantic similarity Bert Score and accuracy BLRU-4 of proposed method reach 67.15 and 16.07, respectively, which outperforms the baseline method in terms of syntax, semantics, and answerability. Experiments also demonstrate that the proposed method can effectively generate first questions adapted to English teaching scenarios.
文摘针对云计算系统在运行过程中由于计算节点空闲而产生大量空闲能耗,以及由于不匹配任务调度而产生大量"奢侈"能耗的能耗浪费问题,提出一种通过任务调度方式的能耗优化管理方法.首先,用排队模型对云计算系统进行建模,分析云计算系统的平均响应时间和平均功率,建立云计算系统的能耗模型.然后提出基于大服务强度和小执行能耗的任务调度策略,分别针对空闲能耗和"奢侈"能耗进行优化控制.基于该调度策略,设计满足性能约束的最小期望执行能耗调度算法ME3PC(minimum expectation execution energy with performance constraints).实验结果表明,该算法在保证执行性能的前提下,可大幅度降低云计算系统的能耗开销.
文摘矩阵主特征向量(principal eigenvectors computing,PEC)的求解是科学与工程计算中的一个重要问题。随着图形处理单元通用计算(general-purpose computing on graphics pro cessing unit,GPGPU)的兴起,利用GPU来优化大规模稀疏矩阵的图形处理单元求解得到了广泛关注。分别从应用特征和GPU体系结构特征两方面分析了PEC运算的性能瓶颈,提出了一种面向GPU的稀疏矩阵存储格式——GPU-ELL和一个针对GPU的线程优化映射策略,并设计了相应的PEC优化执行算法。在ATI HD Radeon5850上的实验结果表明,相对于传统CPU,该方案获得了最多200倍左右的加速,相对于已有GPU上的实现,也获得了2倍的加速。
