Preparation of ZrB_2-SiC Powders via Carbothermal Reduction of Zircon and Prediction of Product Composition by Back-Propagation Artificial Neural Network

Phase pure ZrB2-SiC composite powders were prepared after 1 450℃/3 h via carbothermal reduction route,by using ZrSiO4,B2O3 and carbon as the raw materials.The influences of firing temperature as well as the type and amount of additive on the phase composition of final products were detailedly investigated.The results indicated that the onset formation temperature of ZrB2-SiC was reduced to 1 400℃by the present conditions,and oxide additive（including CoSO4·7H2O,Y2O3 and TiO2）was effective in enhancing the decomposition of raw ZrSiO4,therefore accelerating the synthesis of ZrB2-SiC.Moreover,microstructural observation showed that the as-prepared ZrB2 and SiC respectively had well-defined hexagonal columnar and fibrous morphology.Furthermore,the methodology of back-propagation artificial neural networks（BP-ANNs）was adopted to establish a model for predicting the reaction extent（e g,the content of ZrB2-SiC in final product）in terms of various processing conditions.The results predicted by the as-established BP-ANNs model matched well with that of testing experiment（with a mean square error in 10^（-3） degree）,verifying good effectiveness of the proposed strategy.

神经网络与组合语义在文本相似度中的应用

为了更好地提高短文本语义相似度分析能力,提出了基于神经网络和组合语义的短文本语义相似度分析算法。利用神经网络构建词义表示模型,结合局部和全局上下文信息学习词语在实际语境下的表示;通过句法分析,得到文本的依存关系,并依此构建组合关系树,使用组合语义模型得到整个文本的语义表示;通过计算两个语义表示结果之间的相似性来计算文本的语义相似度。实验分析结果表明,该方法能够在一定程度上提高文本语义分析能力。

基于CNN特征空间的微博多标签情感分类

面对微博情感评测任务中的多标签分类问题时,基于向量空间模型的传统文本特征表示方法难以提供有效的语义特征。基于深度学习的词向量表示技术,能够很好地体现词语的语法和语义关系,且可以依据语义合成原理有效地构建句子的特征表示向量。作者提出一个针对微博句子的多标签情感分类系统,首先从1个大规模的无标注微博文本数据集中学习中文词语的词向量表示,然后采用卷积神经网络(convolution neural network,CNN)模型进行有监督的多情感分类学习,利用学习到的CNN模型将微博句子中的词向量合成为句子向量,最后将这些句子向量作为特征训练多标签分类器,完成微博的多标签情感分类。2013年NLPCC(Natural Language Processing and Chinese Computing)会议的微博情感评测公开数据集中,相比最优评测结果的宽松指标和严格指标,本系统的最佳分类性能分别提升了19.16%和17.75%;采用Recursive Neural Tensor Network模型合成句子向量的方法,取得目前已知文献中的最佳分类性能,系统将2个指标分别提升了3.66%和2.89%。采用多种多标签分类器来对比不同的特征表示方法,发现基于CNN特征空间的句子向量具有最好的情感语义区分度;通过对CNN迭代训练过程的分析,体现了语义合成过程中的模式识别规律。进一步的工作包括引入更多合适的深度学习模型,并深入探索基于词向量的语义合成现象。

Strategies of ＂Power-off Reset＂ Failure for Siemens MM440 in the Application of Gas Press

Gas pressurized fan speed control and remote monitoring was achieved based on three sets of gas pressure MM440 frequency converter control system, through the Profibus-DP network transmission of exchange data and signal instructions. In operation, one converter which adopted ＂ power-off reset＂ has broken down, so as to cause two other gas pressurized frequency converter tripping. This paper proposed method to achieve capture and process failures, and application to converter control system. The result shows that the present method solves the fault, improves the system operation stability and meets the requirement of production.

基于双语合成语义的翻译相似度模型

提出基于双语合成语义的翻译相似度模型,通过在翻译过程中引入双语语义相似度特征提高翻译性能。首先利用分布式方法分别在源端和目标端获取短语的单语合成语义向量,然后利用神经网络将它们映射到同一语义空间,获得双语合成语义向量。在该语义空间,计算源语言短语和对应的目标语言短语之间基于合成语义向量的翻译相似度,将其作为一个新特征加入解码器。在汉英翻译NIST06和NIST08测试数据集上,相较于基准系统,基于双语合成语义的翻译相似度模型获得0.56和0.42 BLEU值的显著性提高。

Influence of conductive network composite structure on the electromechanical performance of ionic electroactive polymer actuators

The influence of the nanostructure of the conductive network composite(CNC)on the performance of ionic electroactive polymer(IEAP)actuators has been examined in detail.We have studied IEAP actuators consisting of CNCs with different volume densities of gold nanoparticles(AuNPs)and the polymer network.Varying the concentration of AuNPs in CNC thin films was used as a means to control the CNC-ion interfacial area and the electrical resistance of the CNC,with minimum effect on the mechanical properties of the actuator.Increasing the interfacial area and reducing the resistance,while maintaining porosity of the composite,provide means for generating motion of more ions into the CNC at a significantly shorter time,which results in generation of strain at a faster rate.We have demonstrated that cationic strain in actuators with denser CNCs is improved by more than 460%.Denser CNC structures have larger interfacial areas,which results in attraction/repulsion of more ions in a shorter time,thus generation of a larger mechanical strain at a faster rate.Also,time-dependent response to a square-wave voltage was improved by increasing the AuNP concentration in the CNC.Under 0.1 Hz frequency,the cationic strain was increased by 64%when the AuNP concentration was increased from 4 to 20 ppm.

A Luminescent Dicyanodistyrylbenzene-based Liquid Crystal Polymer Network for Photochemically Patterned Photonic Composite Film

A novel photonic composite film based on a luminescent dicyanodistyrylbenzene-based liquid crystal polymer network（LCN） was fabricated by using a silica colloidal crystal as a template. The upper part of inverse opal structure and the luminescence characteristics of dicyanodistyrylbenzene-based moiety endowed the resulting bilayer photonic film with structural color arising from coherent Bragg reflection and fluorescence properties, respectively. A fluorescence enhancement phenomenon was observed in the photonic film due to the overlap between the reflection band and emission band of the fluorescent LCN. More importantly, the photo-induced irreversible Z/E photoisomerization of dicyanodistyrylbenzene-based moiety in the photonic film led to both a reflection spectral shift and an observable fluorescence variation. On the basis of this effective phototuning process, microscopic patterning of photonic film was developed under both fluorescence mode and reflection mode. The work demonstrated here provides a new route to construct photo-responsive photonic film.