随着电子病历(EHR)的广泛应用,基于深度学习的临床健康事件预测引起了众多研究者的关注。现有工作主要集中在挖掘患者的高阶时间特征,未能有效地学习疾病之间的隐关系。针对疾病表征学习的问题,本文提出一种新的疾病表示模型(Health Eve...随着电子病历(EHR)的广泛应用,基于深度学习的临床健康事件预测引起了众多研究者的关注。现有工作主要集中在挖掘患者的高阶时间特征,未能有效地学习疾病之间的隐关系。针对疾病表征学习的问题,本文提出一种新的疾病表示模型(Health Event Prediction Model Based on Dynamic and Static Features of Graph Nodes,DuDas)。该模型最终挖掘出的疾病隐表征包含静态和动态信息,最终实现对临床任务的预测。首先根据疾病共现频率构建疾病关系图,并通过one-hot编码模块为每个疾病节点分配一个初始隐表征。然后根据静态挖掘模块挖掘疾病的静态表征,并与相应的初始隐表征融合为初始动态隐表征。根据图卷积模块挖掘疾病之间的动态关系,学习疾病节点的最终动态隐表征。由于患者的就诊记录具有时间性,本文使用门控循环单元来挖掘历史诊断信息与当前诊断信息之间的关系。为了验证本文提出的方法的有效性,在2个真实数据集上进行实验。实验结果表明,本文提出的模型在预测健康事件任务上达到了更高水平。展开更多
This paper deals with an innovative low-loss AC switch, named as TBBS （transistor based bidirectional switch）, based on the association of super-gain BJTs developed by the GREMAN laboratory. The main characterizatio...This paper deals with an innovative low-loss AC switch, named as TBBS （transistor based bidirectional switch）, based on the association of super-gain BJTs developed by the GREMAN laboratory. The main characterization results of the super-gain BJT are reminded to identify the key parameters that are essential to build the TBBS. A complete characterization database in static mode of this new AC switch is discussed. In particular, its forward and reverse-biased features have been measured to see the evolution of the DC current gain as a function of the current density. The TBBS makes sense when using the super-gain BJT （bipolar junction transistor） in reverse mode. It means that the reverse DC current gain has to be sufficient （at least higher than l compared with the conventional BJT one）. This new AC switch is bidirectional in current and voltage, totally controllable （turn-on and turn-off） and the most attractive solution in terms of on-state power losses. Further, its manufacturing process is as easier as existing device such as triac.展开更多
文摘随着电子病历(EHR)的广泛应用,基于深度学习的临床健康事件预测引起了众多研究者的关注。现有工作主要集中在挖掘患者的高阶时间特征,未能有效地学习疾病之间的隐关系。针对疾病表征学习的问题,本文提出一种新的疾病表示模型(Health Event Prediction Model Based on Dynamic and Static Features of Graph Nodes,DuDas)。该模型最终挖掘出的疾病隐表征包含静态和动态信息,最终实现对临床任务的预测。首先根据疾病共现频率构建疾病关系图,并通过one-hot编码模块为每个疾病节点分配一个初始隐表征。然后根据静态挖掘模块挖掘疾病的静态表征,并与相应的初始隐表征融合为初始动态隐表征。根据图卷积模块挖掘疾病之间的动态关系,学习疾病节点的最终动态隐表征。由于患者的就诊记录具有时间性,本文使用门控循环单元来挖掘历史诊断信息与当前诊断信息之间的关系。为了验证本文提出的方法的有效性,在2个真实数据集上进行实验。实验结果表明,本文提出的模型在预测健康事件任务上达到了更高水平。
文摘This paper deals with an innovative low-loss AC switch, named as TBBS （transistor based bidirectional switch）, based on the association of super-gain BJTs developed by the GREMAN laboratory. The main characterization results of the super-gain BJT are reminded to identify the key parameters that are essential to build the TBBS. A complete characterization database in static mode of this new AC switch is discussed. In particular, its forward and reverse-biased features have been measured to see the evolution of the DC current gain as a function of the current density. The TBBS makes sense when using the super-gain BJT （bipolar junction transistor） in reverse mode. It means that the reverse DC current gain has to be sufficient （at least higher than l compared with the conventional BJT one）. This new AC switch is bidirectional in current and voltage, totally controllable （turn-on and turn-off） and the most attractive solution in terms of on-state power losses. Further, its manufacturing process is as easier as existing device such as triac.
