基于门控联合池化自编码器的通用性文本表征

为了学习文本的语义表征,以往的研究者主要依赖于复杂的循环神经网络(recurrent neural networks, RNNs)和监督式学习方法。该文提出了一种门控联合池化自编码器(gated mean-max AAE)用于学习中英文的文本语义表征。该文的自编码器完全通过多头自注意力机制(multi-head self-attention mechanism)来构建编码器和解码器网络。在编码阶段,提出了均值—最大化(mean-max)联合表征策略,即同时运用平均池化(mean pooling)和最大池化(max pooling)操作来捕获输入文本中多样性的语义信息。为促使联合池化表征可以全面地指导重构过程,解码器采用门控操作进行动态关注。通过在大规模中英文未标注语料上训练模型,获得了高质量的句子编码器。在重构文本段落的实验中,该文模型在实验效果和计算效率上均超越了传统的RNNs模型。将公开训练好的文本编码器,使其可以方便地运用于后续的研究。     

2020年4月全国主要橡胶原材料价格

2020年4月全国主要橡胶原材料价格     

古建筑木结构透榫节点力学模型研究

为研究古建筑木结构透榫节点的M-θ力学模型，在分析透榫节点构造特征与受力机理的基础上，建立其数值模型，用透榫节点的试验数据验证了该数值模型的正确性，并分析了节点缝隙、木材横纹弹性模量和大榫头长度对透榫节点受弯承载力的影响。根据受力分析结果，建立以弹性点、屈服点与极限点为特征点的三折线多参数M-θ力学模型，其结果与多数的试验结果基本吻合，并将该力学模型应用于木构架的受力分析。研究结果表明：透榫节点的滞回耗能能力强，节点的变形主要集中在榫头处。当榫头与卯口之间的缝隙增大时，节点的受弯承载力降低。随木材横纹弹性模量的提高和大榫头长度的增加，节点的受弯承载力有一定提高。文章建立的M-θ力学模型能较好反映透榫节点的受力过程，适用于木构架的受力分析，其荷载 位移骨架曲线与试验结果基本吻合。研究成果可为古建筑木结构的维修与保护提供参考。     

非常规蛋白基木材胶黏剂研究现状及发展前景

介绍了动物和植物型非常规蛋白资源的种类及其制备蛋白基木材胶黏剂的研究现状和存在的问题,展望了非常规蛋白胶黏剂的发展前景。     

Mechanophysical Cues in Extracellular Matrix Regulation of Cell Behavior

The extracellular matrix (ECM) is a macromolecular network that can provide biochemical and structural support for cell adhesion and formation. It regulates cell behavior by influencing biochemical and physical cues. It is a dynamic structure whose components are modified, degraded, or deposited during connective tissue development, giving tissues strength and structural integrity. The physical properties of the natural ECM environment control the design of naturally or synthetically derived biomaterials to guide cell function in tissue engineering. Tissue engineering is an important field that explores physical cues of the ECM to produce new viable tissue for medical applications, such as in organ transplant and organ recovery. Understanding how the ECM exerts physical effects on cell behavior, when cells are seeded in synthetic ECM scaffolds, is of utmost importance. Herein we review recent findings in this area that report on cell behaviors in a variety of ECMs with different physical properties, i.e., topology, geometry, dimensionality, stiffness, and tension.     

Microwave dielectric properties of SrLa[Ga1−x(Mg0.5Ti0.5)x]O4 and SrLa[Ga1−x(Zn0.5Ti0.5)x]O4 (x = 0.2-0.8) ceramics

SrLa[Ga_1−x(R_0.5Ti_0.5)_x]O₄ (R = Mg, Zn) ceramics were prepared by a standard solid state sintering method. The single-phase ceramics with K₂NiF₄-type layered perovskite structure and I4/mmm space group were obtained, indicating that SrLa(R_0.5Ti_0.5) and SrLaGaO₄ can form the unlimited solid solutions. With increasing x for R = Mg and Zn, ε_r increases monotonously, the Qf value first increases and then decreases, while τ_f increases from a negative to a positive value. The optimized microwave dielectric properties were obtained as following: ε_r = 23.3, Qf = 89 400 GHz, τ_f = −0.8 ppm/°C for SrLa[Ga_0.6(Mg_0.5Ti_0.5)_0.4]O₄ and ε_r = 23.3, Qf = 76 200 GHz, τ_f = 0.2 ppm/°C for SrLa[Ga_0.7(Zn_0.5Ti_0.5)_0.3]O₄, indicating that the present solid solution ceramics are the promising candidates as microwave resonator materials for the telecommunication applications.     

深度学习识别光网络单元故障的设计与应用

为解决依赖装维上门鉴别光网络单元故障带来的不便,可以从机器视觉入手实现自动化故障识别。近年,ImageNet挑战赛的成功推动了物体识别技术的跨越式发展,特别是基于卷积的深度学习技术在视觉识别方面已经达到人类水平,为光网络单元故障的自动识别提供了技术基础。文章对识别光网络单元的工作状态进行了研究,将设备工作状态分为7个场景,提出了利用手机APP采集图片识别故障的解决方案并投入了实际生产;重点阐述了深度学习模块的设计与实现,提出一种通过算法整合的方式综合运用物体检测和图像分类算法,分3阶段逐步求精,解决了图片过滤,光网络单元型号和状态识别等问题,实现了基于计算机视觉自动识别光网络单元故障。从数据上看产品的端到端准确率超过84%,识别速度达到10 FPS,月均提供服务超过1万人次,在减少用户等待的同时节约了人力资源。