Task-based teaching has been introduced to many EFL educational settings and has become widely used within those contexts.As a new English teaching approach,it has been evaluated by educators.This essay discusses task...Task-based teaching has been introduced to many EFL educational settings and has become widely used within those contexts.As a new English teaching approach,it has been evaluated by educators.This essay discusses task-based language teaching within the context of EFL students in a middle school in China.A discription of the context,problems of introducing this approach in this context is demonstrated,followed by an overview of this approach and an evaluation including its advantages and disadvantages weighed by scholars,as well as their opinions on"tasks"and"grammar".The end of this essay provides a discussion of solutions for the problems of using this ap proach within this context.展开更多
机器阅读理解(MRC)是自然语言处理领域的一个具有挑战性的任务,其目标是在给定文章中预测出相关问题的答案.随着深度学习和预训练语言模型的发展,许多端到端的机器阅读理解模型展现出优秀的性能,但是这些模型普遍存在鲁棒性不足的问题,...机器阅读理解(MRC)是自然语言处理领域的一个具有挑战性的任务,其目标是在给定文章中预测出相关问题的答案.随着深度学习和预训练语言模型的发展,许多端到端的机器阅读理解模型展现出优秀的性能,但是这些模型普遍存在鲁棒性不足的问题,当文本中存在干扰句时,它们的表现便显著下降.本文从人类做阅读理解任务的角度来解决这个问题,提出了一种端到端的多任务学习框架ASMI(Answer-Span Context Prediction and Mutual Information Estimation and Maximization)来提高MRC模型的鲁棒性.ASMI在预训练语言模型下游微调,包含两种辅助任务:(i)答案上下文预测;(ii)答案与上下文之间的互信息估计.本文设计了一种上下文注意力机制来预测答案上下文软标签,从而强化上下文对于问答任务的指导作用,并降低干扰句对模型的影响.本文还提出了一种新的负样本生成策略,并结合基于JS散度的互信息估计器来估计互信息,从而有效辨析答案上下文和干扰句之间的语义差异,使得模型学习到更加鲁棒的表示.在3个阅读理解基准数据集上的实验结果表明,本文提出的ASMI模型在EM和F1指标上均优于对比模型.展开更多
针对现有基于角色访问控制的缺陷和分布式工作流管理系统的特性,在传统的基于角色的访问控制模型中引入任务集(Tasks)、任务实例集(TaskInstances)和任务上下文(TaskContext)的概念,将传统的user role permission权限赋予结构修改为user...针对现有基于角色访问控制的缺陷和分布式工作流管理系统的特性,在传统的基于角色的访问控制模型中引入任务集(Tasks)、任务实例集(TaskInstances)和任务上下文(TaskContext)的概念,将传统的user role permission权限赋予结构修改为user role task permission权限赋予结构,建立了基于任务和角色的访问控制模型,给出了其形式化定义。该模型解决了传统的基于角色访问控制中的动态适应性差和最小权限约束假象的问题,用于分布式工作流管理系统,提高了安全性、实用性。展开更多
文摘Task-based teaching has been introduced to many EFL educational settings and has become widely used within those contexts.As a new English teaching approach,it has been evaluated by educators.This essay discusses task-based language teaching within the context of EFL students in a middle school in China.A discription of the context,problems of introducing this approach in this context is demonstrated,followed by an overview of this approach and an evaluation including its advantages and disadvantages weighed by scholars,as well as their opinions on"tasks"and"grammar".The end of this essay provides a discussion of solutions for the problems of using this ap proach within this context.
文摘机器阅读理解(MRC)是自然语言处理领域的一个具有挑战性的任务,其目标是在给定文章中预测出相关问题的答案.随着深度学习和预训练语言模型的发展,许多端到端的机器阅读理解模型展现出优秀的性能,但是这些模型普遍存在鲁棒性不足的问题,当文本中存在干扰句时,它们的表现便显著下降.本文从人类做阅读理解任务的角度来解决这个问题,提出了一种端到端的多任务学习框架ASMI(Answer-Span Context Prediction and Mutual Information Estimation and Maximization)来提高MRC模型的鲁棒性.ASMI在预训练语言模型下游微调,包含两种辅助任务:(i)答案上下文预测;(ii)答案与上下文之间的互信息估计.本文设计了一种上下文注意力机制来预测答案上下文软标签,从而强化上下文对于问答任务的指导作用,并降低干扰句对模型的影响.本文还提出了一种新的负样本生成策略,并结合基于JS散度的互信息估计器来估计互信息,从而有效辨析答案上下文和干扰句之间的语义差异,使得模型学习到更加鲁棒的表示.在3个阅读理解基准数据集上的实验结果表明,本文提出的ASMI模型在EM和F1指标上均优于对比模型.
文摘针对现有基于角色访问控制的缺陷和分布式工作流管理系统的特性,在传统的基于角色的访问控制模型中引入任务集(Tasks)、任务实例集(TaskInstances)和任务上下文(TaskContext)的概念,将传统的user role permission权限赋予结构修改为user role task permission权限赋予结构,建立了基于任务和角色的访问控制模型,给出了其形式化定义。该模型解决了传统的基于角色访问控制中的动态适应性差和最小权限约束假象的问题,用于分布式工作流管理系统,提高了安全性、实用性。