基于深度记忆网络的特定目标情感分类

只基于注意力机制的深度记忆网络不能有效处理目标上下文情感依赖于具体目标的情况,为了解决该问题,提出了一个对目标敏感的深度记忆网络模型。该模型利用注意力机制来获取决定目标上下文情感的信息,然后通过交互模块将上下文情感表示和上下文与目标之间的交互信息融合成分类特征,最后分类得到目标的情感极性。在SemEval 2014 task4的两个数据集上进行实验,实现了比只基于注意力机制的DMN模型明显更好的◢F▼◣▽1值。实验结果表明,在解决上下文情感依赖于具体目标的问题时,考虑上下文与目标之间的交互信息是有效的。     

图像分割中MRF与DMN方法评述与比较

综述马尔科夫随机场(Markov Random Field,MRF)的研究和应用历史,着重讨论了图像分割中MRF的原理和应用。分析了可分解马尔科夫网(Decomposable Markov Networks,DMN)的一般方法以及DMN在图像分割问题中的应用。比较研究了MRF和DMN的区别和联系。     

Effect of Surface Hydride,Vinyl, and Methyl Groups on Thermal Stability of Modified Silica-Divinylbenzene-Di(Methacryloyloxymethyl)Naphthalene Composites

Divinylbenzene (DVB)-di(methacryloyloxymethyl)naphthalene (DMN)/modified silica composites were studied by means of IR spectroscopy, thermogravimetric analysis, and differential scanning calorimetry. The results show that filling with modified fumed silica decreases the initial temperature of DVB-DMN copolymer thermal-oxidative degradation; however, the activation energy of this process generally increases. The partial modification of silica surface turned out to be more efficient than the total functionalization and silicas containing surface methyl in combination with hydride groups were found to be the best fillers for DVB-DMN copolymer among all those investigated.     

新的动态记忆网络的视觉问答

视觉问答任务旨在给机器输入一幅图像和一相关问题,计算机能够准确作答。针对这一任务,对记忆和注意力机制的神经网络结构进行了深入研究,这类网络显示出问题回答所需的某些推理能力。在分析动态记忆网络的基础上,提出了一种新的动态记忆网络,对原来的DMN的内存和输入模块进行改进。结合这些变化,一个新的图像输入模块引入到视觉问答系统中。在DAQUAR-ALL、COCO-QA和VQA数据集上验证了该方法的有效性。实验结果表明,所提出的新动态记忆模型取得了很好的结果,比一些经典深度方法都更出色。     

From BPMN process models to DMN decision models

The interplay between process and decision models plays a crucial role in business process management, as decisions may be based on running processes and affect process outcomes. Often process models include decisions that are encoded through process control flow structures and data flow elements, thus reducing process model maintainability. The Decision Model and Notation (DMN) was proposed to achieve separation of concerns and to possibly complement the Business Process Model and Notation (BPMN) for designing decisions related to process models. Nevertheless, deriving decision models from process models remains challenging, especially when the same data underlie both process and decision models. In this paper, we explore how and to which extent the data modeled in BPMN processes and used for decision-making may be represented in the corresponding DMN decision models. To this end, we identify a set of patterns that capture possible representations of data in BPMN processes and that can be used to guide the derivation of decision models related to existing process models. Throughout the paper we refer to real-world healthcare processes to show the applicability of the proposed approach.     

啤酒瓶用共缩聚PET/PEN的研究

采用对苯二甲酸二甲酯(DMT)与2,6-萘二甲酸二甲酯(DMN)合成了聚对苯二甲酸二乙酯(PET)和聚2,6-萘二甲酸二乙酯(PEN)共缩聚物,研究了PEN用量对共缩聚PET／PEN热性能的影响,并研究了采用PET／PEN制成的啤酒瓶的物理性能、气体阻隔性能和贮藏性。结果表明：PET／PEN的性能与普通聚酯切片的相近,PEN质量分数为30％的PET／PEN制成的啤酒瓶可耐95℃高温,对O2、CO2气体的阻隔性比普通PET瓶提高6倍。     

聚萘二甲酸乙二醇酯（PEN）生产技术进展

对PEN最新生产技术及由邻二甲苯和丁二烯为原料合成PEN的技术经济性作了综述。基于PEN的发展与其单体价格密切相关,指出降低2,6-DMN生产成本仍将是今后PEN生产技术改进的方向。     

PEN合成的酯交换反应工艺及动力学研究

讨论了使用 (CH3COO) 2 Zn·2H2 O作为催化剂 ,2 ,6 萘二甲酸二甲酯 (DMN) 乙二醇 (EG)的酯交换反应。该酯交换反应遵循二级反应动力学。反应活化能为 14 7.9kJ mol。较高的反应活化能使得该酯交换反应速率对反应温度比较敏感。     

2,6-二甲基萘的分离与提纯方法

介绍了10种二甲基萘异构体的物性特征，并对吸附法、共熔结晶法、高压结晶法、直接结晶法及乳化结晶法分离2，6－二甲基萘的技术进行了综述。