在机器阅读理解任务中,如何在包含不可回答问题的情况下提高答案的准确性是自然语言处理领域的一项重要挑战.虽然基于深度学习的机器阅读理解模型展现出很好的性能,但是这些模型仍然存在抽取特征冗余、语义信息不全面、问题分类任务和...在机器阅读理解任务中,如何在包含不可回答问题的情况下提高答案的准确性是自然语言处理领域的一项重要挑战.虽然基于深度学习的机器阅读理解模型展现出很好的性能,但是这些模型仍然存在抽取特征冗余、语义信息不全面、问题分类任务和答案抽取任务耦合性不强的问题.为了解决以上问题,本文提出一种结合门控机制和多级残差结构的多任务联合训练模型GMRT(Gated Mechanism and Multi-level Residual Structure for Multi-task Joint Training),以提升机器阅读理解任务中答案预测的准确性.GMRT构建门控机制来筛选交互后的关联特征,从而控制信息的流动.采用多级残差结构分别连接注意力机制和门控机制,保证每个阶段都保留原始语义信息.同时,通过边缘损失函数对问题分类任务和答案抽取任务联合训练,确保预测答案过程中任务之间的强耦合性.在SQuAD2.0数据集上的实验结果表明,GMRT模型的EM值和F1值均优于对比模型.展开更多
Different from limb rehabilitation training,the purpose of muscle strength training is to reduce muscle atrophy and increase muscle strength and tolerance through strength training of limb muscles,and then improve the...Different from limb rehabilitation training,the purpose of muscle strength training is to reduce muscle atrophy and increase muscle strength and tolerance through strength training of limb muscles,and then improve the muscle strength level of muscles(groups),mainly for sports fitness and muscle strengthening groups and patients with muscle atrophy or muscle weakness caused by various diseases.In this paper,we developed a new reconfigurable muscle strength training robot,a bionic robot by imitating physicians to conduct muscle strength training for patients,which was developed with six training modes for 17 joint movements,that is,the shoulder flexion/extension,the shoulder internal/external rotation,the shoulder adduction/abduction,the elbow flexion/extension,the wrist supination/pronation,the wrist flexion/extension,the wrist radial/ulnar deviation,the hip flexion/extension,the hip internal/external rotation,the hip adduction/abduction,the knee flexion/extension,the ankle dorsiflexion/plantarflexion,the ankle adduction/abduction,the ankle inversion/eversion,the waist flexion/extension,the waist left/right rotation,and the waist left/right flexion.The reconfigurable mechanism was designed with fully electric adjuster and reconfigurable adaptors deployed on the driving unit,and six training modes were developed,namely,continuous passive motion,active exercise,passive–active exercise,isotonic exercise,isometric exercise and isokinetic exercise.Experiments with knee joint and elbow joint have shown that the developed reconfigurable muscle strength training robot can realize the multi-mode trainings for the 17 joint movements.展开更多
文摘在机器阅读理解任务中,如何在包含不可回答问题的情况下提高答案的准确性是自然语言处理领域的一项重要挑战.虽然基于深度学习的机器阅读理解模型展现出很好的性能,但是这些模型仍然存在抽取特征冗余、语义信息不全面、问题分类任务和答案抽取任务耦合性不强的问题.为了解决以上问题,本文提出一种结合门控机制和多级残差结构的多任务联合训练模型GMRT(Gated Mechanism and Multi-level Residual Structure for Multi-task Joint Training),以提升机器阅读理解任务中答案预测的准确性.GMRT构建门控机制来筛选交互后的关联特征,从而控制信息的流动.采用多级残差结构分别连接注意力机制和门控机制,保证每个阶段都保留原始语义信息.同时,通过边缘损失函数对问题分类任务和答案抽取任务联合训练,确保预测答案过程中任务之间的强耦合性.在SQuAD2.0数据集上的实验结果表明,GMRT模型的EM值和F1值均优于对比模型.
基金supported in part by the National Key R&D Program of China(No.2018YFB1307004)in part by the National Natural Science Foundation of China(Nos.61903011 and 52175001)。
文摘Different from limb rehabilitation training,the purpose of muscle strength training is to reduce muscle atrophy and increase muscle strength and tolerance through strength training of limb muscles,and then improve the muscle strength level of muscles(groups),mainly for sports fitness and muscle strengthening groups and patients with muscle atrophy or muscle weakness caused by various diseases.In this paper,we developed a new reconfigurable muscle strength training robot,a bionic robot by imitating physicians to conduct muscle strength training for patients,which was developed with six training modes for 17 joint movements,that is,the shoulder flexion/extension,the shoulder internal/external rotation,the shoulder adduction/abduction,the elbow flexion/extension,the wrist supination/pronation,the wrist flexion/extension,the wrist radial/ulnar deviation,the hip flexion/extension,the hip internal/external rotation,the hip adduction/abduction,the knee flexion/extension,the ankle dorsiflexion/plantarflexion,the ankle adduction/abduction,the ankle inversion/eversion,the waist flexion/extension,the waist left/right rotation,and the waist left/right flexion.The reconfigurable mechanism was designed with fully electric adjuster and reconfigurable adaptors deployed on the driving unit,and six training modes were developed,namely,continuous passive motion,active exercise,passive–active exercise,isotonic exercise,isometric exercise and isokinetic exercise.Experiments with knee joint and elbow joint have shown that the developed reconfigurable muscle strength training robot can realize the multi-mode trainings for the 17 joint movements.
