基于度量学习的意图识别和槽填充方法

人机对话中小样本学习场景下的意图识别和槽填充,是自然语言处理的一个重要课题.本文采用基于度量学习的方法,通过计算query set中的样本与support set中样本的距离,寻找距离最近的类别样本作为分类标签,同时将两个任务联合进行训练,用以提升模型的效果.从实验结果中可以得出,本文提出的Fine-tune方法,对意图识别和槽填充任务都有一定的帮助和提升.胶囊网络在意图识别中也起到了一定的效果,可以帮助去除一部分无关信息,但对槽填充任务的帮助不明显;而任务自适应的投影网络,可以更好地将不同类的向量分开,提升了两个任务的性能.     

喹啉类杂环化合物的设计与合成研究

通过Vilsmeier反应合成了2-氯-3-甲酰基喹啉中间体,利用该化合物设计合成了一系列新的多杂原子稠环化合物,并通过核磁氢谱、红外光谱、元素分析对它们进行了结构表征。     

渣油催化临氢热转化钼铁双金属催化剂的研究

Replacement of precious single metal catalysts with cost-effective, highly-dispersed composite catalysts for catalytic hydrothermal conversion of residue holds tremendous promise for the residue upgrading technologies. Organic metals were added to the feed as the oil-soluble precursors, and transformed into the catalytic active phases in this work. Physical properties and structures of the composite catalysts had been investigated by X-ray fluorescence spectroscopy, X-ray photoelectron spectroscopy, X-ray diffraction, scanning electron microscope and transmission electron microscopy. The composite catalysts were found to be highly efficient in the catalytic hydrothermal conversion of both model compound and residue. Increased metal dispersion and synergistic effects of two metals played indispensable roles in such catalytic system. Results showed that under the test conditions in the article, the catalyst had the best catalytic performance when the mass ratio of molybdenum to iron was 1.5.     

绝缘栅双极晶体管表面电沉积镍基石墨烯薄膜的制备及其散热性能研究

将镍基石墨烯薄膜应用在IGBT模块散热方面,提出了镍基石墨烯薄膜的制备工艺,并利用扫描电镜、透射电镜、能谱仪等对镍基石墨烯薄膜进行表征。结果表明,当电流密度为2 A/dm~2时,镍基石墨烯薄膜中的有大量的Ni和C元素,镍基石墨烯薄膜中的石墨烯平均粒径为32.5 nm。当脉冲电流密度为1 A/dm~2时,镍基石墨烯薄转移镍基石墨烯薄膜后的S1芯片的散热效果较好。当脉冲电流密度为2 A/dm~2时,镍基石墨烯薄转移镍基石墨烯薄膜后的S2芯片的散热效果最好,其最大散热量为14.9℃。     

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。     

基于小样本下统计不确定性的边坡可靠度Bootstrap-BC_a方法

传统方法对小样本下的岩土抗剪强度参数估计区间较宽,会导致工程设计偏于保守,为此,提出一种考虑测量数据统计不确定性的边坡可靠度分析Bootstrap-BC_a法。该方法在Bootstrap的基础上加入偏差修正量■和加速量a,进一步校正了置信区间的端点值,提高了估计精度。在保证样本容量和变异性一致的条件下,对比分析Bootstrap-BC_a法和Bootstrap法在95%置信水平下的置信区间以及相对应的安全系数估计值。结果表明:Bootstrap-BC_a法能够有效收窄区间宽度且保证率下限值更趋近实际值;Bootstrap-BC_a法计算所得边坡安全系数均值相对较大,离散型更小,且与样本容量成正比,与安全系数方差成反比;3种变异水平的安全系数均值都随着样本容量n的增大而增大,而安全系数方差则逐渐减小,且大变异水平下计算所得安全系数均值更大。     

Synthesis and Characterization of Pr0.6-xNdxCa0.4FeO3-δ with Pechini Method

Pr0.6-x NdxCa0.4 FeO3-δ （ x = 0.0, 0.2, 0.3, 0.4, 0.5, 0.6） were synthesized using Pechini method. A number of studies were conducted concerning composition, specific area, crystalline structure and microstructure of the samples by means of FT-IR, BET, XRD TG-DTA and SEM. The results show that all the samples with different doping amounts of Pr^3＋ and Nd^3＋ on A-site are fine dispersed, and mean particle size less than 100 nm. The powders have good sinterability, and the relative density is 95% after sintered at 1200 ℃ for 2 h. It is found that all specimens are entirely single phase solid solutions with orthorhombic perovskite structure, the stable perovskitetype phase is formed completely after calcination at 900 ℃.